Long-term Outcomes Associated With Implantable Cardioverter Defibrillator in Adults With Chronic Kidney Disease

Key Points

Question  Does placement of implantable cardioverter defibrillators improve clinical outcomes in patients with chronic kidney disease?

Findings  In this cohort study of 5877 community-based patients with heart failure and chronic kidney disease, use of implantable cardioverter defibrillators was not significantly associated with improved survival but was associated with increased risk for subsequent heart failure and all-cause hospitalization.

Meaning  The potential risks and benefits of implantable cardioverter defibrillators should be carefully considered in patients with heart failure and chronic kidney disease.

Importance  Chronic kidney disease (CKD) is common in adults with heart failure and is associated with an increased risk of sudden cardiac death. Randomized trials of participants without CKD have demonstrated that implantable cardioverter defibrillators (ICDs) decrease the risk of arrhythmic death in selected patients with reduced left ventricular ejection fraction (LVEF) heart failure. However, whether ICDs improve clinical outcomes in patients with CKD is not well elucidated.

Objective  To examine the association of primary prevention ICDs with risk of death and hospitalization in a community-based population of potentially ICD-eligible patients who had heart failure with reduced LVEF and CKD.

Design, Settings, and Participants  This noninterventional cohort study included adults with heart failure and an LVEF of 40% or less and measures of serum creatinine levels available from January 1, 2005, through December 31, 2012, who were enrolled in 4 Kaiser Permanente health care delivery systems. Chronic kidney disease was defined as an estimated glomerular filtration rate of less than 60 mL/min/1.73 m². Patients who received and did not receive an ICD were matched (1:3) on CKD status, age, and high-dimensional propensity score to receive an ICD. Follow-up was completed on December 31, 2013. Data were analyzed from 2015 to 2017.

Exposures  Placement of an ICD.

Main Outcomes and Measures  All-cause death, hospitalizations due to heart failure, and any-cause hospitalizations.

Results  A total of 5877 matched eligible adults with CKD (1556 with an ICD and 4321 without an ICD) were identified (4049 men [68.9%] and 1828 women [31.1%]; mean [SD] age, 72.9 [8.2] years). In models adjusted for demographics, comorbidity, and cardiovascular medication use, no difference was found in all-cause mortality between patients with CKD in the ICD vs non-ICD groups (adjusted hazard ratio, 0.96; 95% CI, 0.87-1.06). However, ICD placement was associated with increased risk of subsequent hospitalization due to heart failure (adjusted relative risk, 1.49; 95% CI, 1.33-1.60) and any-cause hospitalization (adjusted relative risk, 1.25; 95% CI, 1.20-1.30) among patients with CKD.

Conclusions and Relevance  In a large, contemporary, noninterventional study of community-based patients with heart failure and CKD, ICD placement was not significantly associated with improved survival but was associated with increased risk for subsequent hospitalization due to heart failure and all-cause hospitalization. The potential risks and benefits of ICDs should be carefully considered in patients with heart failure and CKD.

Chronic kidney disease (CKD)¹ is a major public health condition that is estimated to affect 14% of US adults.² Cardiovascular disease is the leading cause of morbidity and death in patients with CKD,³ with heart failure being one of the most common cardiovascular manifestations.^4-6 Concurrently, more than 5.7 million adults in the United States are estimated to have heart failure, of whom 30% have CKD.^7,8 A complication of heart failure is sudden cardiac death, with CKD being one of the strongest risk factors for this outcome.⁹

Placement of an implantable cardioverter defibrillator (ICD) as a primary prevention strategy for sudden cardiac death reduced the risk of death due to arrhythmia in adults with heart failure and reduced left ventricular ejection fraction (LVEF) compared with optimal medical therapy alone in selected participants enrolled in randomized trials.^10-12 However, patients with CKD are notably underrepresented in existing trials. Because ICD placement carries risks and is expensive, a better understanding of how best to use this therapy in high-risk subgroups such as patients with CKD is critical.

Existing studies of primary prevention ICDs in patients with CKD have been limited by modest sample size,^9,13-15 highly selected populations of trial participants^9,16 (who are typically healthier and less representative compared with target patients with heart failure¹⁷), and lack of a comparison group of similar patients with CKD who did not undergo ICD placement.^13,18 Therefore, in this noninterventional study, we examined the association of primary prevention ICD placement with the risk of death and hospitalization in a community-based population of potentially ICD-eligible patients with reduced LVEF heart failure and CKD.

The study population included members from 4 participating health care delivery systems within the Cardiovascular Research Network from January 1, 2005, through December 31, 2013.¹⁹ Sites included Kaiser Permanente Northern California, Kaiser Permanente Southern California, Kaiser Permanente Northwest, and Kaiser Permanente Colorado. Each site has a virtual data warehouse that served as the primary data source for participant identification and characterization.²⁰ The virtual data warehouse is a distributed, standardized data resource that consists of electronic data sets at each site that are populated with linked demographic, administrative, outpatient pharmacy, laboratory testing, and health care resource (ambulatory visits and network and nonnetwork hospitalizations with diagnoses and procedures) data. Institutional review boards at Kaiser Permanente Northern California, Kaiser Permanente Southern California, Kaiser Permanente Northwest, and Kaiser Permanente Colorado approved the study and waived the need for informed consent owing to the retrospective nature of the study.

We identified adults 21 years or older who had at least 12 months of continuous health plan enrollment and pharmacy benefit before the index date to ensure adequate data on covariates. Patients had a diagnosis of heart failure based on having been hospitalized with a primary discharge diagnosis of heart failure and/or having at least 3 ambulatory visits coded for heart failure, with at least 1 visit being with a cardiologist, from January 1, 2005, through December 31, 2012.²¹ We used the following codes from the International Classification of Diseases, Ninth Revision (ICD-9): 398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 428.0, 428.1, 428.20, 428.21, 428.22, 428.23, 428.30, 428.31, 428.32, 428.33, 428.40, 428.41, 428.42, 428.43, and 428.9. Previous studies have shown a positive predictive value of greater than 95% for admissions with a primary discharge diagnosis of heart failure based on these codes compared against manual medical record review using Framingham clinical criteria.^22,23

Because we were interested in patients who were eligible for ICD placement, we further ascertained information on quantitative and/or qualitative assessments of left ventricular systolic function from the results of echocardiograms, radionuclide scintigraphy, other nuclear imaging modalities, and left ventriculography test results available from site-specific databases complemented by manual medical record review.²¹ Only patients with an LVEF of 40% or less or moderate to severely reduced systolic function on qualitative assessment with no history of ICD placement were included (Figure 1).^12,24 We only included patients with CKD, which was defined²⁵ as an estimated glomerular filtration rate (eGFR) of less than 60 mL/min/1.73 m² by the Chronic Kidney Disease Epidemiology Collaboration equation¹ using the closest outpatient serum creatinine level to 365 days before or on the index date. We excluded patients who had missing measures of serum creatinine levels at study entry, had preexisting end-stage renal disease (defined as receipt of long-term dialysis or a kidney transplant), did not have follow-up data, or had a previous solid organ transplant (Figure 1).

The main exposure of interest was ICD placement during follow-up. Placement of an ICD was identified based on any 1 of the following procedure codes: 00.51, 00.52, 00.54, 37.94, 37.95, 37.96, 37.97, 37.98, 37.99, V45.02, 37.75, 37.79, 37.99, or 89.49 from ICD-9; 00534, 33215, 33216, 33217, 33218, 33220, 33224, 33225, 00534, 33240, 33241, 33243, 33244, 33249, 33223, 33230, 33231, 33248, 33262, 33263, 33264, 93283, 93284, 93287, 93737, 93738, or 93745 from Current Procedural Terminology; and C1721, C1722, C1777, C1882, C1895, C1896, or C1899 from the Health Care Financing Administration Common Procedural Coding System. Patients were also required to have an outpatient, non–emergency department serum creatinine value in the 1 year before ICD placement, no history of end-stage renal disease, no history of organ transplantation, and available follow-up after the ICD placement date. With these criteria, 3312 patients who received ICDs were eligible for matching (Figure 1).

Patients were censored during follow-up at the time of any solid organ transplant or the end of the follow-up period on December 31, 2013. We studied the following 3 primary outcomes: all-cause mortality, heart failure–related hospitalizations, and any-cause hospitalizations. All-cause mortality was identified through health system administrative databases, member proxy reporting, hospitalizations, regional cancer registries, Social Security Administration vital status files, and state death certificate files.^26,27 Heart failure–related hospitalizations were identified using a principal discharge diagnosis of heart failure based on validated ICD-9 diagnosis codes 398.91, 402.x1, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, and 428.x.²¹ Hospitalizations for any cause were identified through Kaiser Permanente’s comprehensive hospitalization and billing claims databases for network and nonnetwork admissions, respectively.

We ascertained information on demographic characteristics and comorbidity based on previously validated diagnoses and procedures using ICD-9 codes, laboratory results, prescribed medications, and registries.³ Potential confounders included demographic characteristics (age, sex, and race/ethnicity), comorbid conditions based on relevant diagnosis and procedure codes (prior acute coronary syndrome, coronary revascularization, ischemic stroke or transient ischemic attack, atrial fibrillation, ventricular fibrillation or tachycardia, mitral or aortic valvular heart disease, obstructive sleep apnea, peripheral arterial disease, tobacco use, dyslipidemia, hypertension, diabetes, dementia, depression, lung disease, liver disease, and cancer), ambulatory measures of blood pressure, body mass index, low- and high-density lipoprotein cholesterol levels, hemoglobin level, LVEF, and selected medication use (angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, β-blockers, and diuretics) within 120 days before the index date from pharmacy dispensing databases.

Data were analyzed from 2015 to 2017. All analyses were conducted using SAS statistical software (version 9.3; SAS Institute). We conducted a matched parallel cohort study to examine the association between ICD placement and subsequent outcomes of death, hospitalization for heart failure, and any-cause hospitalization. The cohort entry date was the date of ICD placement in the ICD group. The same calendar date was also the cohort entry date for the matched non-ICD group. We used a 2-step process for matching. We first estimated a high-dimensional propensity score²⁸ for receipt of ICD for each patient.²⁸ We developed the high-dimensional propensity score using techniques by Schneeweiss et al.^28-30 In brief, the high-dimensional propensity score algorithm (1) required the identification of approximately 500 variables from different data dimensions (eg, demographic, hospitalization, procedure, outpatient care, laboratory, and medication data) in the database, (2) identified the most prevalent variables in each data dimension as candidate covariates, (3) ranked candidate covariates based on their occurrence (the frequency that the codes were recorded for each individual during the baseline period), (4) ranked covariates across all data dimensions by their potential for control of confounding based on the bivariate associations of each covariate with the treatment and with the outcome, (5) selected covariates from step 4 (eg, 200) for propensity score modeling, and (6) estimated the propensity score with multivariable logistic regression using the selected covariates. We then individually matched patients who did or did not receive an ICD based on age (±5 years), being alive on the ICD placement date (or the same calendar date for the non-ICD group), CKD status (eGFR <60 or ≥60 mL/min/1.73 m²) as of the matching date, and having a high-dimensional propensity score difference of 0.005 or less. In addition, matched patients who did not receive an ICD also were required to have at least 365 days of continuous health plan enrollment and pharmacy benefits before the corresponding match date, no history of end-stage renal disease or organ transplant, and an outpatient serum creatinine level measured within 365 days before the match date. We chose this approach for the non-ICD group to avoid immortal time bias.^31-33 On the basis of these criteria, we successfully matched 1556 unique patients who underwent ICD with CKD and 4321 patients with CKD who did not receive an ICD (Figure 1).

Characteristics of matched patients were compared using analysis of variance or a relevant nonparametric test for continuous variables and χ² tests for categorical variables. Given the large sample size, standard differences in each variable were compared between matched groups by computing a difference in means of the 2 groups divided by the pooled SD, with D values of greater than 0.10 considered to be statistically significant. Rates of each outcome (per 100 person-years) with associated 95% CIs were calculated for matched patients. After confirming no violation in the proportional hazards assumption, Cox proportional hazards regression models were used to examine the association between receipt of a primary prevention ICD and risk of all-cause death, and a generalized estimating equation Poisson regression with robust SEs was used to examine the association of ICD placement with heart failure–related hospitalizations and any-cause hospitalizations to allow multiple events per patient. We performed a series of sequential, nested models using the following approach. Model 1 adjusted for age, sex, and race. Model 2 adjusted for the covariates in model 1 and added baseline smoking, prevalent heart failure, acute myocardial infarction, unstable angina, coronary bypass surgery, percutaneous coronary intervention, ischemic stroke or transient ischemic attack, other arterial thromboembolic event, atrial fibrillation or flutter, ventricular tachycardia or fibrillation, mitral and/or aortic valvular disease, peripheral artery disease, rheumatic heart disease, pacemaker, dyslipidemia, hypertension, diabetes, diagnosed dementia, diagnosed depression, chronic lung disease, chronic liver disease, systemic cancer, LVEF, hemoglobin level, systolic blood pressure, high- and low-density lipoprotein cholesterol levels, eGFR, and calendar year of study entry. Finally, model 3 adjusted for covariates in model 2 and added baseline use of targeted cardiopreventive medications (angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, β-blockers, and diuretics).

Among the 5877 matched patients with reduced LVEF heart failure and CKD (4049 men [68.9%] and 1828 women [31.1%]; mean [SD] age, 72.9 [8.2] years; 1556 with ICD and 4321 without ICD), those with ICD placement had lower mean (SD) LVEF (26.8% [7.7%] vs 29.4% [8.2%]), were more likely to have known coronary heart disease (577 [37.1%] vs 1301 [30.1%]), and were more likely to be taking a loop diuretic (1288 [82.8%] vs 3174 [73.5%]). However, the ICD group was less likely to have diagnosed dementia (43 [2.8%] vs 166 [3.8%]) (Table 1). Overall, the patients were well-matched on all other covariates.

Overall, 2541 patients (43.2%) died during a mean (SD) follow-up of 3.1 (2.3) years. The crude rate of death was 14.9 per 100 person-years (95% CI, 13.9-16.1 per 100 person-years) in the ICD group vs 13.6 per 100 person-years (95% CI, 13.0-14.2 per 100 person-years) in the non-ICD group, with no significant difference in the unadjusted probability of survival during follow-up (Table 2 and Figure 2A). Furthermore, the association between receipt of an ICD and all-cause death was not statistically significant in adjusted models (Table 2).

A total of 1922 patients (32.7%) were hospitalized for heart failure during follow-up. Crude rates of heart failure–related hospitalization were significantly higher in the ICD group (16.90 per 100 person-years; 95% CI, 15.64-18.27 person-years) than in the non-ICD group (11.12 per 100 person-years; 95% CI, 10.53-11.74 per 100 person-years) (Table 3 and Figure 2B). In Poisson models, ICD implantation was associated with a higher risk of heart failure–related hospitalization, even after adjustment for demographics, comorbidities, and medication use (adjusted relative risk, 1.49; 95% CI, 1.39-1.60) among patients with CKD (Table 3).

In the study population, a total of 4093 patients (69.6%) were hospitalized for any cause during follow-up. The crude rate of hospitalization was significantly higher in the ICD group (51.65 per 100 person-years; 95% CI, 48.79-54.68 per 100 person-years) compared with the non-ICD group (38.27 per 100 person-years; 95% CI, 36.91-39.69 per 100 person-years) (Table 3 and Figure 2C). In models adjusted for demographics, comorbidities, and medication use, ICD placement was associated with a 25% higher relative risk of hospitalization for any cause (adjusted relative risk, 1.25; 95% CI, 1.20-1.30) compared with receiving no ICD implant (Table 3) among patients with rEF heart failure and CKD.

In this study of nearly 6000 contemporary matched patients with CKD and reduced LVEF heart failure who were potentially eligible for a primary prevention ICD, we found that after accounting for a broad range of potential confounders, ICD placement was not significantly associated with a lower risk of all-cause death but was associated with higher adjusted rates of heart failure–related and any-cause hospitalization. These noninterventional study findings suggest that ICDs do not provide a survival benefit in a contemporary population with CKD with reduced LVEF heart failure and may be associated with greater morbidity (eg, hospitalizations). In the absence of randomized trials of ICD therapy that include a large number of patients with moderate to advanced CKD, these data may offer insight in weighing the net potential risks and benefits to make clinical decisions regarding ICD placement in current practice among patients with CKD, who constitute a notable proportion of patients with reduced LVEF heart failure in the United States.

In our study, ICD placement was not significantly associated with a lower risk of death in patients with reduced LVEF heart failure and CKD after careful matching and further adjustment for other potential confounding factors. Few studies have examined the effect of CKD on outcomes among patients with heart failure who are eligible for an ICD. The limited studies that have included patients with kidney disease have largely relied on administrative diagnostic codes to define CKD (which is known to lead to misclassification) or have been limited to patients undergoing long-term dialysis,³⁴ who represent a substantially smaller, higher-risk subset of the larger population with CKD. Several studies have reported that among patients who receive an ICD, those with CKD have a higher risk of death. Among Medicare beneficiaries, CKD was associated with a 2.3-fold higher rate of death after ICD implantation.¹⁸ In 958 patients with heart failure who received a primary prevention ICD, a stepwise increase in mortality for every worsening stage of CKD was found.¹³ In a single-center study of 199 patients with CKD who received an ICD,¹⁴ more advanced CKD was also associated with higher mortality. These prior studies have been limited only to patients who received an ICD (without a comparison group of patients with CKD who did not receive an ICD).

Only a few studies^15,34 have compared outcomes in patients with heart failure and CKD who did vs did not receive an ICD. For example, in a prospective study of 900 patients eligible for ICD,³⁴ investigators found no difference in the association of ICD with all-cause death in the 23 patients with kidney failure who received dialysis vs all other patients. In a study of 108 patients with eGFR less than 30 mL/min/1.73 m² who received a primary prevention ICD matched to patients without an ICD,¹⁵ prophylactic ICD placement also did not confer a survival advantage. Although informative, these prior studies are limited by small sample size and did not account for a broad range of confounders, including laboratory data and receipt of cardiovascular medications. Another strength of our study is the systematic classification of CKD, which was defined by ambulatory, non–emergency department eGFR measures proximal to ICD placement rather than administrative diagnostic codes.

Hospitalizations are an important outcome to patients as a quality of life measure and pose substantial economic burdens to the health care system. Patients with CKD are known to have a disproportionate burden of hospitalizations and recurrent hospitalizations even without placement of cardiac devices.^35,36 In our analyses, we found that ICD placement was independently associated with greater risks of heart failure–related hospitalization and any-cause hospitalization in patients with CKD and reduced LVEF heart failure. Our findings extend data from other studies in non-CKD populations.^37,38 In a single-center study of 180 patients with an ICD,³⁷ more than half of patients were rehospitalized within the first 1 to 2 years. In a retrospective analysis of the Multicenter Automatic Defibrillator Implantation Trial (MADIT) II trial,³⁸ investigators found that participants who received an ICD had a 90% greater risk of heart failure–related hospitalization compared with those randomly assigned to conventional therapy, with a similar effect size as observed in our study. Our study expands on these prior studies by examining a large, contemporary, and representative population with well-characterized reduced LVEF heart failure and CKD. Although ICD placement may reduce risk of arrhythmic events in certain groups, patients with heart failure remain at high risk for recurrent hospitalizations. Furthermore, ICDs have known complications (eg, ICD device infections) that may contribute to higher rates of hospitalization overall.

The findings of this noninterventional study may have important therapeutic implications, particularly given the paucity of clinical trial data related to ICD placement in patients with CKD. These data call for a more comprehensive view of the net risks and benefits of ICD placement in eligible patients with reduced LVEF heart failure and CKD and for future trials to help directly address these questions. The findings are consistent with the literature on patients with end-stage renal disease undergoing dialysis that has suggested that patients may be too sick to benefit from primary prevention ICD placement owing to high competing risks of death^39,40 and post-ICD complications.^39,41,42 Similarly, a recent analysis of patients with continuous-flow left ventricular assist devices also noted that the presence of an ICD was not associated with improved survival.⁴³

Our study had several strengths. We studied a large, contemporary, multicenter, well-characterized population of patients in typical clinical practice. We used advanced statistical methods to account for confounding, including high-dimensional propensity score matching and additional adjustment for a broad range of possible confounders, including detailed laboratory data and receipt of cardiovascular medications. Chronic kidney disease was defined using ambulatory measures of eGFR proximal to the index date. We also recognize certain limitations. The outcome of heart failure–related hospitalization was ascertained using relevant ICD-9 diagnosis codes. However, previous work^22,23,44-46 has validated the high positive predictive value of this approach for this outcome. Specific outcome data on arrhythmic events or death and cause of hospitalization were not systematically available. Health care intensity and surveillance (which is often subjective and clinician and patient dependent) are possible confounders in this analysis; however, these are not easily obtained comprehensively through electronic health data. Our study population largely included patients with moderate stages of CKD with reduced LVEF heart failure; further work is needed to determine whether these findings are generalizable to other populations with CKD. In our study population, 1878 (32.0%) had known coronary heart disease; thus, our findings may not be generalizable to all patients with reduced LVEF heart failure, particularly those with ischemic heart failure. Despite individual matching on key confounders and additional statistical adjustment for a wide range of other potential explanatory factors, we cannot completely exclude residual or unmeasured confounding or selection bias,⁴⁷ which may have affected our findings. The findings from our study should be confirmed with future randomized clinical trials. Patients in the study were enrolled in health care delivery systems and thus may not be completely generalizable to all uninsured patient populations. Finally, we were not able to determine causality in this noninterventional study.

Placement of ICDs was not independently associated with lower all-cause mortality but was associated with higher adjusted risks of heart failure–related and any-cause hospitalization in a large, community-based population of adults with moderate CKD and reduced LVEF heart failure. The risks and benefits should be carefully balanced in the decision to place an ICD in patients with CKD and heart failure.

Accepted for Publication: December 11, 2017.

Corresponding Author: Nisha Bansal, MD, MAS, Kidney Research Institute, Division of Nephrology, University of Washington, 908 Jefferson St, 3rd Floor, Seattle, WA 98104 (nbansal@nephrology.washington.edu).

Published Online: February 5, 2018. doi:10.1001/jamainternmed.2017.8462

Author Contributions: Ms Tabada and Dr Go had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Bansal, Smith, Gurwitz, Masoudi, Greenlee, Dighe, Go.

Acquisition, analysis, or interpretation of data: Bansal, Szpiro, Reynolds, Gurwitz, Masoudi, Greenlee, Tabada, Sung, Go.

Drafting of the manuscript: Bansal, Tabada, Dighe.

Critical revision of the manuscript for important intellectual content: Bansal, Szpiro, Reynolds, Smith, Gurwitz, Masoudi, Greenlee, Tabada, Sung, Go.

Statistical analysis: Bansal, Szpiro, Tabada.

Obtained funding: Bansal, Masoudi, Go.

Administrative, technical, or material support: Masoudi, Greenlee, Sung, Dighe.

Study supervision: Reynolds, Smith, Go.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported grants R56HL121069 and U19 HL91179-01 from the National Heart, Lung, and Blood Institute, grant K23 DK088865 from the National Institute of Diabetes and Digestive and Kidney Diseases, and grant HHSA290-2005-0033-I-TO8-WA from the US Department of Health and Human Services.

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank all the project managers, data programmers, and analysts for their critical technical contributions and support that made this study possible.