Trends, etiologies, and predictors of 90‐day readmission after percutaneous ventricular assist device implantation: A national population‐based cohort study

Abstract

Percutaneous ventricular assist devices (pVADs) are indicated to provide hemodynamic support in high‐risk percutaneous interventions and cardiogenic shock. However, there is a paucity of published data regarding the etiologies and predictors of 90‐day readmissions following pVAD use. We studied the data from the US Nationwide Readmissions Database (NRD) for the years 2013 and 2014. Patients with a primary discharge diagnosis of pVAD use were collected by searching the database for International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) procedural code 37.68 (Impella and TandemHeart devices). Amongst this group, we examined 90‐day readmission rates. Comorbidities as identified by “CM_” variables provided by the NRD were also extracted. The Charlson Comorbidity Index was calculated using appropriate ICD‐9‐CM codes, as a secondary diagnosis. A 2‐level hierarchical logistic regression model was then used to identify predictors of 90‐day readmission following pVAD use. Records from 7074 patients requiring pVAD support during hospitalization showed that 1562 (22%) patients were readmitted within 90 days. Acute decompensated heart failure (22.6%) and acute coronary syndromes (11.2%) were the most common etiologies and heart failure (odds ratio [OR]: 1.39, 95% confidence interval [CI]: 1.17–1.67), chronic obstructive pulmonary disease (OR: 1.26, 95% CI: 1.07–1.49), peripheral vascular disease (OR: 1.305, 95% CI: 1.09–1.56), and discharge into short‐ or long‐term facility (OR: 1.28, 95% CI: 1.08–1.51) were independently associated with an increased risk of 90‐day readmission following pVAD use. This study identifies important etiologies and predictors of short‐term readmission in this high‐risk patient group that can be used for risk stratification, optimizing discharge, and healthcare transition decisions.

1. INTRODUCTION

Over the last decade, mechanical circulatory support (MCS) devices are increasingly used as an integral component in the management of patients with cardiogenic shock, decompensated heart failure (HF), and patients undergoing complex percutaneous coronary intervention (PCI).1 Cardiogenic shock complicates 5% to 8% of patients presenting with acute ST‐segment elevation myocardial infarction.2 Although intra‐aortic balloon pump (IABP) failed to show any overall mortality benefit,3 hemodynamic support with the use of percutaneous MCS has been shown to reduce morbidity and mortality in these high‐risk patients.4, 5 Among commonly used temporary percutaneous MCS devices, also known as percutaneous ventricular assist devices (pVADs), are TandemHeart, Impella 2.5, Impella CP, and Impella 5.6, 7 These devices can be implanted percutaneously, obviating the need for surgery, as in the case of left ventricular assist devices.6 Whereas the methods of implantation, hemodynamic effects, and pump mechanics of these two devices differ, they work on a common principle of maintaining perfusion, decompressing the left ventricle, and minimizing oxygen requirement in the left ventricle.8 pVADs have consistently demonstrated far superior hemodynamic profiles compared to IABPs in all trials, which albeit did not translate into survival benefit.1, 9, 10 Furthermore, with increasing utilization of pVADs, hospitalizations associated with them and their complications have a significant impact on healthcare utilizations and cost. This is the first national population‐based study assessing etiologies, trends, and predictors of 90‐day readmissions after the use of a pVAD. This would enable us to possibly recognize potentially vulnerable areas for future intervention to prevent readmissions.

2. METHODS

The study cohort was derived from the Healthcare Cost and Utilization Project's National Readmission Database (NRD) from 2013 to 2014, sponsored by the Agency for Healthcare Research and Quality. The NRD is one of the largest publicly available all‐payer inpatient‐care databases in the United States, including data on approximately 15 million discharges in years 2013 to 2014, estimating roughly 35 million discharges from 22 states with reliable, verified linkage numbers. NRD represents 49.3% of total US hospitalizations. Patients were tracked during the same year using the variable NRD VisitLink, and time between 2 admissions was calculated by subtracting the variable NRD_DaysToEvent. The time to readmission was calculated by subtracting the length of stay of index admission to the time between 2 admissions. National estimates were produced using sampling weights provided by the sponsor. The details regarding the NRD data are available online.11

We queried NRD database using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) procedural code for insertion of percutaneous external heart assist device 37.68. The patients with age < 18 years and those missing data for age, gender, or mortality were excluded. We also excluded index admissions in the months of October, November, and December as we did not have 90‐day follow‐up data for the same. We identified a total of 7074 index admissions. Similar methods have also been used previously.11, 12 The patients who were readmitted to the hospital within 90 days within the same calendar year were further evaluated (n = 1562).

The primary outcome of our study was 90‐day readmissions, and secondary outcomes were etiologies and predictors of readmission. Causes of readmission were identified by using ICD‐9‐CM codes in primary diagnosis filed during readmission observation. We identified 282 different ICD‐9‐CM diagnosis codes and combined the ones with similar diagnoses to make clinically important groups.

NRD variables were used to identify patients' demographic characteristics such as age and gender, hospital features including bed size and teaching status, and patient‐specific characteristics consisting of median household income category for the patient's zip code, primary payer, admission type, admission day, and discharge disposition.13 Comorbidities such as obesity, hypertension, diabetes mellitus, HF, chronic pulmonary disease, vascular disease history, anemia, neurological disease or paralysis, hematological or oncological malignancy, and renal failure were identified by variables provided in the NRD,14 which uses ICD‐9‐CM diagnoses and the diagnosis‐related group in effect on the discharge date, and it is not associated with primary diagnosis.

Severity of comorbid conditions was defined using Deyo modification of Charlson Comorbidity Index (CCI), which contains comorbid conditions with differential weights. The score ranges from 0 to 33, with greater scores corresponding to greater burden of comorbid diseases.15, 16, 17

SAS 9.4 (SAS Institute Inc., Cary, NC) was utilized for analysis. Differences between categorical variables were tested using the χ² test, and differences between continuous variables were verified using the Student t test. Hierarchal 2‐level logistic regression model with hospital identifiation as the random effect was used to evaluate secondary outcomes. For 90‐day multivariate readmission model, we included hospital level characteristics such as hospital teaching status and bed size; patient's demographics such as age and gender; comorbid conditions such as obesity, hypertension, diabetes mellitus, HF, chronic pulmonary disease, peripheral vascular disease, anemia, and renal failure; and other patient level characteristics such as primary payer, median household income, and disposition after hospital discharge. P values of <0.05 were considered significant.

3. RESULTS

We identified 7074 patients admitted for pVAD, of whom 1562 (22%) patients were readmitted within 90 days. A majority of patients were male (74%), age 50 to 80 years (72%), and had significant baseline burden of comorbidities with CCI score (≥2 [77.1%]. HF (74.8%), hypertension (60.2%), diabetes mellitus (38.6%), and chronic pulmonary disease (20.5%) were the most common comorbidities. Overall, 68.2% had Medicare/Medicaid as the primary payer, 80.7% had it as a nonelective procedure, and 34.5% of the patients died during the index hospitalization. Average hospital length of stay (LOS) of index admission was 11.8 ±0.1 days, whereas 49.5% of patients were discharged to home, and 15.8 were discharged to a facility (Table 1).

Table 1.

Baseline characteristics

Variables	90‐Day Readmission		Overall	P Value
	No	Yes
Primary admission	5512 (78%)	1562 (22.09%)	7074 (100%)
Patient level variables
Age, y				0.007
18–49	12.38%	11.36%	12.15%
50–64	33.16%	30.12%	32.49%
65–79	39.31%	40.29%	39.53%
≥80	15.15%	18.23%	15.83%
Gender				0.460
Male	5220 (73.8%)	5286 (74.73%)	5235 (74%)
Female	1853 (26.2%)	1788 (25.27%)	1839 (26%)
Charlson/Deyo Comorbidity Indexa				<0.001
0	3.26%	2.92%	3.19%
1	20.94%	15.38%	19.71%
≥2	75.79%	81.7%	77.1%
Comorbiditiesb
Obesity	15.69%	13.64%	15.2%	0.0304
History of hypertension	59.22%	63.74%	60.22%	0.001
Diabetes mellitus	37.92%	41.05%	38.61%	0.025
Chronic pulmonary disease	19.06%	25.67%	20.52%	<0.001
Peripheral vascular disease	16.99%	22.27%	18.15%	<0.001
Peripheral vascular disease	16.99%	22.27%	18.15%	<0.001
Neurological disorder or paralysis	2.27%	2.09%	2.23%	0.679
Anemia	20.94%	23.52%	21.51%	0.028
Hematological or oncological malignancy	2.72%	2.9%	2.76%	0.7053
Weight loss	4.8%	5.1%	4.9%	0.096
Median household income category for patients' zip code (percentile)c				0.780
0–25th percentile	29.49%	28.94%	29.37%
26th–50th percentile	27.7%	29.05%	28%
51th–75th percentile	23.85%	23.29%	23.73%
76th–100th percentile	18.95%	18.72%	18.9%
Primary payer				<0.001
Medicare/Medicaid	66.13%	75.64%	68.23%
Private including HMO	24.91%	17.23%	23.21%
Self‐pay/no charge/other	8.96%	7.02%	8.53%
Hospital characteristics
Hospital bed sizec				0.725
Small	5.23%	5.48%	5.29%
Medium	18.53%	17.71%	18.53%
Large	76.24%	76.81%	76.36%
Hospital teaching status				0.030
Nonteaching	25.48%	22.82%	24.89%
Teaching	74.52%	77.12%	75.11%
Admission type				0.600
Nonelective	80.83%	8.23%	80.7%
Elective	19.17%	19.77%	19.3%
Admission day				0.031
Weekdays	79.34%	81.82%	79.89%
Weekend	20.66%	18.18%	20.11%
Disposition				<0.001
Home	42.8%	72.9%	49.45%
Facility/others	12.69%	26.97%	15.84%
In‐hospital mortality	44.33%	0	34.54%	<0.001
Length of stay, d, mean	11.53	10.69	11.18	<0.001

Abbreviations: HMO: health maintenance organization.

^aCharlson/Deyo Comorbidity Index was calculated as per Deyo classification.

^bVariables are Agency for Healthcare Research and Quality comorbidity (cm_) measures.

^cThe bed‐size cutoff points divided into small, medium, and large have been done so that approximately one‐third of the hospitals in a given region, location, and teaching status combination would fall within each bed‐size category. (State and County QuickFacts. Washington, DC: US Census Bureau; 2012.)

^dRepresents a quartile classification of the estimated median household income of residents in the patient's zip code, derived from zip code demographic data obtained from Claritas. The quartiles are identified by values of 1 to 4, indicating the poorest to wealthiest populations. Because these estimates are updated annually, the value ranges vary by year (http://www.hcupus.ahrq.gov/db/vars/zipinc_qrtl/nisnote.jsp).

A total of 49.1% of readmissions took place within 30 days of the index hospital discharge. The first 25% and 75% readmissions in general took place in 12 days and 58 days, respectively (Figure 1).

Figure 1.

Ninety‐day readmission trends in patients after percutaneous ventricular assist device implantation

The primary etiologies responsible for 90‐day readmissions are summarized in Table 2. Amongst the etiologies of readmission, cardiac causes (52.5%) were the most common, out of which acute decompensated HF (22.6%) was the most common, followed by acute coronary syndromes (ACS) (11.2%) and arrhythmias (6.9%). Pulmonary etiologies included pneumonia (2.02%), respiratory failure (1.1%), and chronic obstructive pulmonary disease (COPD) (1.7%). Bleeding complications were identified in 4.7% of the readmissions. Infections and acute kidney injury (AKI) were responsible for 9.1% and 6.5% of the readmissions, respectively (Figure 2).

Table 2.

Etiologies of 90‐day readmission

Etiologies	Results, % (N)
Primary admission	100 (7074)
Readmission within 90 days	22 (1562)
No readmission within 90 days	78 (5512)
Cardiac etiology	52.53
Heart failurea	22.65
Arrhythmias	6.97
Atrial fibrillation	0.46
Atrial flutter	0.64
Coronary artery disease/myocardial infarction	11.19
Heart valve disease	2.47
Hypertension	0.18
Hypotension/dizziness/syncope	2.47
Pericardial diseases	0.56
Vascular etiology	4.18
Infections (excluding Pneumonia)	10.4
Respiratory etiology	5.69
Respiratory failure	1.1
Chronic obstructive lung disease/Bronchitis	1.74
Pneumonia	2.02
Others	0.83
Gastrointestinal etiologies	3.99
Neurological	1.82
Ischemic stroke/transient ischemic attack	1.1
Kidney/genitourinary etiologies	6.49
Electrolyte/acid base imbalance	0.46
Acute/acute on chronic kidney failure	3.49
Dehydration/volume depletion	0.55
Urinary tract infections/others	1.99
Hematologic	1.55
Trauma/fracture/poisoning	1.45
Endocrine	2.01
Diabetes mellitus	1.74
Malignancy	4.74
Psychiatric conditions	0.63
Othersb	1.9

^aSystolic, diastolic, combined, acute, chronic, unspecified, volume overload.

^bSkin, subcutaneous, joints, nonspecific lab findings, nonspecific symptoms.

Figure 2.

Pie chart of etiologies of 90‐day readmission of patient who were admitted for percutaneous ventricular assist device placement

Multivariate data analysis of 90‐day readmission after pVAD placement is presented in Table 3. Comorbidities such as HF (odds ratio [OR]:1.39, 95% confidence interval [CI]: 1.172–1.671, P = 0.000), chronic pulmonary disease (OR: 1.26, 95% CI: 1.07–1.49, P < 0.006), peripheral vascular disease (OR: 1.30, 95% CI: 1.09–1.56, P = 0.003), and renal failure (OR: 1.18, 95% CI: 1.01–1.39, P < 0.039) were significant predictors of increased readmission. Patients who were discharged to facilities (OR: 1.28, CI: 1.08–1.51, P < 0.004) after their hospitalization were more likely to be readmitted within 90 days of index hospitalization.

Table 3.

Multivariate predictors of 90‐day readmission

Variables	90‐Day Readmission
		95% CI
	Odds Ratio	LL	UL	P Value
Age, y
18–49	Referent	Referent	Referent
50–64	1.137	0.887	1.458	0.312
65–79	1.010	0.779	1.221	0.527
≥80	0.909	0.677	1.221	0.527
Female	0.993	0.677	1.221	0.527
Charlson/Deyo Comorbidity Indexa
1	Referent	Referent	Referent
2	1.08	1.01	1.15	0.021
≥3	1.16	1.08	1.24	<0.001
Comorbiditiesb
Obesity	0.925	0.749	1.143	0.470
Hypertension	0.967	0.826	1.132	0.676
Diabetes mellitus	0.906	0.780	1.052	0.194
Heart failure	1.399	1.172	1.671	0.000
Chronic pulmonary disease	1.264	1.070	1.494	0.006
Peripheral vascular disease	1.305	1.093	1.558	0.003
Renal failure	1.188	1.009	1.399	0.039
Anemia	0.926	0.780	1.100	0.382
Median household income category for patient's zip codec
0–25th percentile	Referent	Referent	Referent
26th–50th percentile	1.03	0.82	1.29	0.821
51st–75th percentile	0.98	0.78	1.25	0.901
76th–100th percentile	0.98	0.76	1.26	0.876
Primary payer
Medicare/Medicaid	Referent	Referent	Referent
Private including HMO	0.532	0.436	0.648	<0.001
Self‐pay/no charge/other	0.700	0.527	0.929	0.014
Hospital bed sized
Small	Referent	Referent	Referent
Medium	1.306	0.766	2.198	0.315
Large	1.340	0.837	2.144	0.223
Hospital teaching status
Nonteaching	Referent	Referent	Referent
Teaching	1.078	0.882	1.316	0.465
Disposition
Home	Referent	Referent	Referent
Facility/others	1.276	1.080	1.507	0.004

Abbreviations: CI, confidence interval; HMO: health maintenance organization; LL, lower limit; UL, upper limit.

^aCharlson/Deyo Comorbidity Index was calculated as per Deyo classification.

^bVariables are Agency for Healthcare Research and Quality comorbidity (cm_) measures.

^cRepresents a quartile classification of the estimated median household income of residents in the patient's zip code, derived from zip code demographic data obtained from Claritas. The quartiles are identified by values of 1 to 4, indicating the poorest to wealthiest populations. Because these estimates are updated annually, the value ranges vary by year (http://www.hcupus.ahrq.gov/db/vars/zipinc_qrtl/nisnote.jsp).

^dThe bed‐size cutoff points divided into small, medium, and large have been done so that approximately one‐third of the hospitals in a given region, location, and teaching status combination would fall within each bed‐size category. (State and County QuickFacts. Washington, DC: US Census Bureau; 2012.)

4. DISCUSSION

To our knowledge, this is the first study that has simultaneously assessed the etiologies, trends, and predictors of readmission in patients who received pVAD. HF, ischemic heart disease, and arrhythmias were among the most common etiologies of 90‐day readmissions. One‐fifth of patients were readmitted in 90 days, and half of these patients were admitted within a month of discharge. HF, COPD, renal failure, and peripheral vascular disease were the most important predictors of readmissions.

As per the American College of Cardiology/Heart Failure Society of America 2015 guidelines, percutaneous MCS, particularly Impella 2.5 and TandemHeart, can provide superior hemodynamic support as compared to pharmacological therapy in patients with cardiogenic shock.7 pVADs reduce myocardial oxygen consumption, improve mean arterial pressure, and reduce pulmonary capillary wedge pressure, leading to a rapid improvement in hemodynamics and maintenance of end‐organ perfusion.7 In 2014, a study by Maini et al. compared the percutaneous cardiac assist devices with surgical hemodynamic support (extracorporeal membrane oxygenation and extracorporeal ventricular assist device), and showed that pVADs were associated with better outcomes, shorter lengths of stay, lower costs, and a survival benefit when compared with traditional surgical hemodynamic support for patients in cardiogenic shock.16 Even though its superiority was established, we still have limited knowledge on the follow‐up of patients with a pVAD postdischarge and their reasons for subsequent hospitalizations. Using contemporary national readmissions data, we demonstrated that 22% patients with pVADs were readmitted to the hospital within 90 days. The most common causes for readmission in the first 90 days after discharge were acute decompensated heart failure (22.65%), ACS (11.19%), and cardiac arrhythmias (6.97%) (Figure 2). In an analysis of National Inpatient Sample Data from 2007 through 2012, HF was listed as a discharge diagnosis in 70.1% patients.17 These patients tend to be at the highest risk for HF readmission; thus, it is not surprising that HF was the most common reason for readmission. In the same analysis, coronary artery disease was listed as a discharge diagnosis in 82.4% cases.17 In our study, many of these patients had multivessel disease in other territories and a high ischemic burden; thus, they would be more likely to present with ACS. Cardiac arrhythmias constituted the third most common reason for admission, which can be explained by both ongoing ischemia and myocardial fibrosis in these patients with adverse left ventricular remodeling.

The Prospective, Multi‐center, Randomized Controlled Trial of the IMPELLA RECOVER LP 2.5 System Versus Intra‐Aortic Balloon Pump (IABP) in Patients Undergoing Non Emergent High Risk PCI (PROTECT II trial) randomized patients with complex 3‐vessel disease or unprotected left main coronary artery disease and severely depressed left ventricular function to intra‐aortic balloon pump or Impella 2.5 support during nonemergent high‐risk PCI, depicting trends for improved outcomes for Impella 2.5–supported patients at 90 days.18 Amongst 226 patients with Impella support in the PROTECT II trial, 40% had a major adverse event in the first 90 days. Increased incidence of adverse events in the 90‐day follow‐up period in the PROTECT II trial in comparison to our data can be explained by the inclusion of relatively sicker patients in the trial. Furthermore, the experience with implantation has improved over the years, which could have played a part in reducing adverse events and hence readmissions in the follow‐up period in the current analysis. Another study compared pVAD and extracorporeal membrane oxygenation in patients with cardiogenic shock showing 90‐day readmission rate to be 38.7% and 53% in patients with pVAD and extracorporeal membrane oxygenation, respectively.19 Assuming a major adverse event would lead to hospitalization, our real‐world data from the national database had a much lower readmission rate of 22% at 90 days.

In our study, presence of comorbid medical conditions, such as HF, peripheral vascular disease, renal failure, and COPD, was independently associated with 90‐day readmission. As discussed above, patients with HF would be expected to have a greater likelihood of readmissions. The presence of peripheral vascular disease is associated with technical challenges in placement of pVADs and potentially higher incidence of limb ischemia, vascular injury, and bleeding episodes requiring blood transfusion, all of which can lead to rehospitalization. In a study of 90 patients receiving Impella 2.5 or Impella‐CP devices at a single center, vascular complications were seen in 17% and were found to increase the LOS and a tendency toward withdrawal of care.20 This study did not find any effect of vascular complications on 30‐day or 1‐year mortality. However, readmissions were not assessed in this study.

Chronic kidney disease is an important comorbidity in patients undergoing pVAD placement and is present in up to a quarter of the cases.17 In addition, patients with ACS and cardiogenic shock may present with AKI due to reduced perfusion and increased filling pressures. In general, pVADs may be expected to improve renal function if the baseline renal dysfunction is due to low cardiac output, poor renal perfusion, or renal venous congestion. In fact, Impella 2.5 pVAD has been shown to have a renal protective effect in patients undergoing high‐risk PCI regardless of the presence of chronic kidney disease or ejection fraction.21 In a study of 8 patients receiving TandemHeart support for severe refractory cardiogenic shock associated with critical aortic valve stenosis, significant improvement was noted in renal function, with improvement in urine output in 6 out of 8 patients who were anuric prior to the initiation of support.22 The development of AKI during an admission requiring pVAD placement is likely a marker of worse hemodynamics, possible right ventricular failure, and progressive cardiorenal syndrome; thus, it may be a marker for rehospitalization. AKI has also been reported from hemolysis in patients with Impella 2.5.23 Another predictor of readmission is COPD, which is fairly common in these patients; it was listed as a comorbid condition in 18.4% of the patients in the National Inpatient Sample Data from 2007 through 2012.17 These patients may be readmitted for COPD exacerbations and hypoxemic respiratory failure, sometimes along with the right HF. Patients with COPD are also less likely to receive adequate doses of β‐blockers, thus adversely affecting reverse remodeling. Additionally, discharge into a short‐ or long‐term facility was also independently associated with readmissions, which is likely because the cohort discharged to a facility was sicker and had worse functional status.

Gregory et al.24 studied the clinical and economic benefits of Impella 2.5 over conventional IABP during a 90‐day episode of care. Nineteen patients with pVAD had higher hospital costs for index admission ($47 667 vs $33 684; P < 0.001). The total 90‐day hospital charges were similar for the pVAD and the IABP ($172 564 vs $172 758, respectively; P = 0.785). Contrary to the results from the PROTECT II trial, Shah et al used 2010 to 2011 Medicare MEDPAR data, and showed that pVADs were not associated with improved clinical outcomes when compared to IABPs in patients with high‐risk PCI and cardiogenic shock. They further concluded that management of high‐risk PCI and cardiogenic shock with an IABP was more cost effective, with savings of up to $2.5 billion annually to the hospital system.25 Irrespective of these diametrically different results, with tremendous resources and higher cost involved in the care of these patients, it is prudent to understand the causes and comorbid medical conditions associated with hospitalizations to prevent and manage them aggressively.

In our study, several limitations should be acknowledged. Although we used data from the well‐designed NRD, there are still a few limitations. Being a national database analysis, limitations of our study included dependence on documentation of diagnoses. In addition, our analysis focused on readmission in first 90 days of index hospitalization. Also, we cannot adjust our results based on the indication of implantation or the duration of pVAD implantation. Whether the pVAD implantation was a bridge to other therapies including left ventricular assist devices or even heart transplantation is unknown. Further studies are needed with longer follow‐up.

5. CONCLUSION

This study showed a 22% readmission rate in the first 90 days postimplantation and identified the main reasons and independent predictors for readmission. With increasing implantation of pVADs for the management of acute cardiogenic shock, further research is needed to identify patients who are at high risk for readmission. It is important for clinicians to understand the complications and identify the risk factors for readmission to intervene early, thereby decreasing its economic burden on healthcare and improving the quality of life of the patient.

Conflicts of interest

The authors declare no potential conflicts of interest.

Virk HUH, Tripathi B, Gupta S, et al. Trends, etiologies, and predictors of 90‐day readmission after percutaneous ventricular assist device implantation: A national population‐based cohort study. Clin Cardiol. 2018;41:561–568. 10.1002/clc.22929