Non–Evidence-Based ICD Implantations in the United States FREE

Several randomized controlled trials have proven the efficacy of the implantable cardioverter-defibrillator (ICD) at preventing sudden cardiac death in patients with advanced systolic heart failure.^{1 - 3} These trials excluded patients who were in the acute phase of a myocardial infarction (MI), had recent coronary revascularization, had New York Heart Association (NYHA) class IV symptoms, or had newly diagnosed heart failure. In other clinical trials, survival benefit from ICD therapy could not be demonstrated in patients recovering from an acute MI and patients who received an ICD at the time of coronary artery bypass graft (CABG) surgery.^{4 - 6}

The 2006⁷ and 2008⁸ practice guidelines for ICD therapy mandate at least a 40-day period following an MI before an ICD is implanted for a primary prevention indication. These guidelines also emphasize that ICD therapy is not indicated for patients with NYHA class IV symptoms, who are not candidates for a cardiac resynchronization therapy device. In addition, these guidelines specify that recommendations for primary prevention ICDs apply only to patients whose left ventricular ejection fraction is low (≤30% or ≤35%) despite receiving optimal medical therapy. Because this criterion is highly unlikely to be met by patients with newly diagnosed heart failure, these guidelines imply that ICD therapy is not recommended for patients with a new diagnosis of heart failure.^{7 - 8} The degree to which physicians in routine clinical practice follow these evidence-based recommendations is not clear.

We analyzed data from the National Cardiovascular Data Registry’s (NCDR’s)ICD Registry (a national registry of ICD implantations) to determine the number and characteristics of patients recovering from an acute MI or CABG surgery, patients with NYHA class IV symptoms, or patients with newly diagnosed heart failure who receive an ICD, to compare the characteristics and in-hospital outcomes of such patients with those of patients who receive an evidence-based ICD, and to examine the distribution of these implants by site, physician specialty, and year of procedure.

When the Centers for Medicare & Medicaid Services announced their expanded coverage for ICD implantation for the primary prevention of sudden cardiac death in January 2005, the agency mandated that data on all such implants in Medicare beneficiaries be entered into a national ICD Registry.⁹ To respond to this mandate, the Heart Rhythm Society partnered with the American College of Cardiology to establish a national registry of ICD implantations, which was launched on June 30, 2005. The registry is funded by hospital fees and grants from device companies and payers. Although the Centers for Medicare & Medicaid Services mandated submission of data only for primary prevention ICDs in Medicare beneficiaries, 78% of the 1448 participating hospitals are submitting data on all ICD implants (including procedures performed on patients not receiving Medicare) and those performed for secondary prevention of sudden cardiac death. Because these participating hospitals are generally larger, these data account for 90% of the more than 520 000 ICD implants entered into the registry as of April 2010.¹⁰

Details of the ICD Registry were published.¹¹ After formal training on data collection and entry by the NCDR, participating hospitals submit data directly to the ICD Registry via a secure Web site. Submitted data undergo rigorous electronic quality checks. If the data do not pass completeness criteria, participating hospitals can clean their data and resubmit as often as needed until the data pass. Hospital sites are randomly selected annually for an onsite audit (10%). Via quarterly reports, the NCDR shares data with participating hospital sites on their rates of approved indications for primary prevention ICD implantations and in-hospital mortality and other adverse events. Their results are displayed in comparison with a national average for each of these end points.¹¹

Because our study predates version 2.1 of the ICD Registry, all data analyzed in this study were collected using version 1 of the data collection form. The ICD Registry was queried to identify adult patients (≥18 years) with ischemic or nonischemic cardiomyopathy who underwent initial ICD implantation for a primary prevention indication between January 1, 2006, and June 30, 2009. Patients were excluded from this analysis if they had an ICD implanted for a secondary prevention indication or for inducible sustained ventricular tachycardia on electrophysiologic testing, received an ICD with cardiac resynchronization therapy, or received device replacements.

All patients included in this analysis had a prior MI and left ventricular ejection fraction of 30% or lower, or prior congestive heart failure and left ventricular ejection fraction of 35% or lower. Patients were classified as receiving a non–evidence-based ICD implant if they met any of the following criteria: (1) had an MI within 40 days before ICD implantation; (2) had CABG surgery within 3 months before ICD implantation; (3) had NYHA class IV symptoms; or (4) had newly diagnosed heart failure at the time of ICD implantation (a patient could meet >1 criterion). Patients were classified as receiving an evidence-based ICD implant if they met none of these criteria.

We determined the number and demographic and clinical characteristics of patients within each of the non–evidence-based ICD implant subgroups. We compared the characteristics and in-hospital outcomes of patients receiving a non–evidence-based ICD implant with those of patients receiving an evidence-based ICD implant. In-hospital outcomes that were examined included death, any postprocedure complication, cardiac tamponade, pneumothorax, infection, hematoma, and length of hospital stay. We also examined the distribution of non–evidence-based ICD implants by site, physician specialty, and year of procedure.

Details of the methods to determine physician certification were published.¹² In brief, the databases of the American Board of Internal Medicine, the Society for Thoracic Surgeons, and the American College of Surgeons were manually searched to determine physician certification using a combination of physician name and either the National Provider Identifier or Unique Physician Identification Number.¹² The categories included electrophysiologists, nonelectrophysiologist cardiologists, thoracic surgeons, and other. Physicians in the other category included internists and surgeons.

This study was approved by the institutional review board of the Duke University Health System, which determined that informed consent was not applicable to data collected by the ICD Registry.

The baseline characteristics of the different subgroups of patients with a non–evidence-based ICD implant are presented as medians and interquartile ranges (IQRs; 25th and 75th percentiles) for continuous variables and as percentages for categorical variables. We compared the characteristics and in-hospital outcomes of patients receiving a non–evidence-based ICD implant with those of patients receiving an evidence-based ICD implant using the Wilcoxon rank sum test for continuous variables due to the nonnormality of variable distributions, by the Kolmogorov-Smirnov test, and by the χ² test for categorical variables. Differences were declared to be statistically significant when they yielded a P value of less than .05. All statistical tests were 2-sided. A logistic regression model was used to adjust in-hospital outcomes for age, sex, atrial fibrillation or flutter, prior ventricular tachycardia, cerebrovascular disease, chronic lung disease, diabetes, end-stage renal disease, and left ventricular ejection fraction. We conducted a sensitivity analysis in which we excluded patients with NYHA class IV symptoms to determine the effect that these patients had on the main results.

Individual sites established the race and ethnicity of patients receiving an ICD and submitted these data to the ICD Registry. Race is included as a data element in all the NCDR registries because of the importance of assessing how access to medical care, selection of specific therapies, and procedural complications may be related to race.

All hospital sites (N = 1227) were included in this study. In the analysis of the distribution of non–evidence-based ICD implants by site, we excluded sites performing fewer than 20 implants (n = 315 with 2786 patients) and an additional 7748 patients whose records were missing site identifiers.

In analyzing the distribution of non–evidence-based ICD implants by physician specialty, we excluded 12 090 records with no data on physician specialty. Using the likelihood ratio χ² tests, we compared rates of non–evidence-based ICD implants between each physician specialty group and used electrophysiologists as the reference group. We examined temporal changes in non–evidence-based ICD implants from January 1, 2006, through June 30, 2009. The Mantel-Haenszel test was used to test for trends over time in the use of non–evidence-based ICD implants as a proportion of all implants in each year (6 months for 2009). Proportions of each of the non–evidence-based ICD implant subgroups were similarly tested. We used SAS version 8.2 (SAS Institute Inc, Cary, North Carolina) for all statistical analyses.

Of the 112 678 patients who met our inclusion criteria, 971 were excluded for missing data. Of the remaining 111 707 initial primary prevention ICD implants that occurred between January 1, 2006, and June 30, 2009, 25 145 were for a non–evidence-based indication (22.5%). Of these 25 145 non–evidence-based ICD implants, 9257 were in patients within 40 days of an MI (36.8%), 814 were in patients within 3 months of CABG surgery (3.2%), 3022 were in patients with NYHA class IV symptoms (12.0%), and 15 604 were in patients with newly diagnosed heart failure (62.1%).

The baseline characteristics of patients receiving any non–evidence-based ICD implant are presented in Table 1. The median age of the patients was 67 years (IQR, 57-75 years). The majority were men (75.4%) and white (77.4%). Most patients had heart failure (91.8%) and ischemic heart disease (77.2%). The median left ventricular ejection fraction was 25% (IQR, 20%-30%). The government was the primary insurance payer for 66% of these patients. In 63.3% of the patients, the ICD was a dual-chamber device.

Patients who received a non–evidence-based ICD were significantly older and had more comorbid disease than patients who received an evidence-based ICD (Table 1). Specifically, patients who received a non–evidence-based ICD were more likely to have heart failure, atrial fibrillation or flutter, ischemic heart disease, cerebrovascular disease, chronic lung disease, diabetes, and end-stage renal disease. In addition, patients who received a non–evidence-based ICD were more likely to belong to a racial minority group (other than black) and to receive a dual-chamber ICD.

The demographic and clinical characteristics of the 4 subgroups of patients who received a non–evidence-based ICD are presented in Table 2. The median age ranged from 64 to 68 years. The majority of patients in all subgroups were male and white. The vast majority of patients had heart failure and ischemic heart disease in 3 of the subgroups. In the subgroup of patients with NYHA class IV symptoms, patients were more likely to have non−ischemic dilated cardiomyopathy than heart failure and ischemic heart disease. The majority of patients received a dual-chamber ICD. Some patients with NYHA class IV symptoms and a QRS duration of greater than 120 ms (n = 869) were potentially eligible for a cardiac resynchronization therapy device, but did not receive one.

The risk of in-hospital death was significantly higher in patients who received a non–evidence-based device than in patients who received an evidence-based device (0.57% [95% confidence interval {CI}, 0.48%-0.66%] vs 0.18% [95% CI, 0.15%-0.20%], respectively; P <.001) (Table 3). Likewise, the risk of any postprocedure complication was significantly higher in the non–evidence-based ICD group at 3.23% (95% CI, 3.01%-3.45%) compared with 2.41% (95% CI, 2.31%-2.51%) in the evidence-based group (P<.001). Hematoma involving the ICD pocket was more common in patients receiving a non–evidence-based device (0.87% [95% CI, 0.76%-0.99%] vs 0.71% [95% CI, 0.65%-0.77%]; P = .009). There was a nonsignificantly increased incidence of device-related infection in the non–evidence-based ICD group (0.04% [95% CI, 0.02%-0.07%] vs 0.02% [95% CI, 0.01%-0.03%]; P = .06). The risk of cardiac tamponade and pneumothorax was not significantly different between the 2 groups (P = .11 and P = .56, respectively).

Adjusting for potential confounders, any adverse event and death were significantly higher in patients who received a non–evidence-based device (P<.001). There was a nonsignificantly increased risk of hematoma in the non–evidence-based ICD group (P = .07). The median length of hospital stay was significantly longer for patients who received a non–evidence-based ICD compared with patients who received an evidence-based ICD (3 days vs 1 day; P <.001). When these analyses were repeated after excluding patients with NYHA class IV symptoms, the rates of any postprocedure complication, death, and hematoma were significantly higher in patients who received a non–evidence-based ICD (Table 3).

There was significant variation in the distribution of non–evidence-based ICD implants across sites with no clustering of such implants at a subset of sites (Figure 1). The proportion of ICD implants performed by the different types of physician specialty was 66 309 (66.6%) for electrophysiologists, 24 706 (24.8%) for nonelectrophysiologist cardiologists, 2561 (2.6%) for thoracic surgeons, and 6041 (6.1%) for other specialists. The rate of non–evidence-based ICD implants was significantly lower for electrophysiologists (20.8%; 95% CI, 20.5%-21.1%) than nonelectrophysiologists (24.8% [95% CI, 24.2%-25.3%] for nonelectrophysiologist cardiologists; 36.1% [95% CI, 34.3%-38.0] for thoracic surgeons; and 24.9% [95% CI, 23.8%-25.9%] for other specialties) (P<.001 for all comparisons).

There was no clear decrease in the rate of non–evidence-based ICDs over time (24.5% [6908/28 233] in 2006, 21.8% [7395/33 965] in 2007, 22.0% [7245/32 960] in 2008, and 21.7% [3597/16 549] in 2009; P <.001 for trend from 2006-2009 and P = .94 for trend from 2007-2009). The only subgroup that showed a significant decrease in non–evidence-based ICD implants over time as a proportion of all implants was patients within 40 days of an MI (10.5% in 2006, 7.7% in 2007, 7.9% in 2008, and 6.6% in 2009; P <.001 for trend from 2006-2009 and from 2007-2009) (Figure 2).

In a national sample of an initial primary prevention ICD implantation in 111 707 recipients, we found an appreciable number (n = 25 145; 22.5%) of non–evidence-based ICD implants (ie, patients who were either excluded from the major primary prevention clinical trials of ICD therapy or shown not to benefit from an ICD in other trials). Patients who received a non–evidence-based ICD had significantly more comorbidities than patients who received an evidence-based device and were at a higher risk of postprocedural complications (including death).

To our knowledge, our study is the first to examine in-hospital outcomes of patients receiving a non–evidence-based ICD nationally. Our findings suggest that 1 excess complication occurred for every 121 non–evidence-based ICD implantations. Although the absolute difference in complications between the 2 groups is modest, these complications could have significant effects on patients' quality of life and health care use, including length of hospital stay and costs. Importantly, these complications resulted from procedures that were not clearly indicated in the first place. While a small risk of complications is acceptable when a procedure has been shown to improve outcomes, no risk is acceptable if a procedure has no demonstrated benefit.

The increased prevalence of comorbidities in recipients of non–evidence-based ICDs is undoubtedly associated with an increased risk of competing causes of death. Compared with patients who received an evidence-based ICD, patients who received a non–evidence-based ICD are more likely to have worse intermediate and long-term outcomes including mortality. However, this finding needs to be confirmed by future studies.

During this period of limited resources and due to the Centers for Medicare & Medicaid Services' emphasis on quality improvement by promoting evidence-based care, it is increasingly important to assess hospital performance and to provide feedback to hospitals about their outcomes and compliance with clinical guideline recommendations. Providing such feedback to hospitals has the potential to improve adherence to practice guidelines and eventually patient outcomes.

In this study, we found substantial hospital variation in the use of non–evidence-based ICDs, which at many sites constituted more than 40% of the overall number of implanted ICDs. Therefore, there is an opportunity to educate hospital sites and physicians to improve their adherence to guidelines. This study highlights the significant role that the ICD Registry could play in quality improvement, including the quality benchmarking reports provided to hospitals comparing their outcomes with a national aggregate.¹³

In contrast to a previous analysis of the ICD Registry in which black and Hispanic patients were significantly more likely than white patients to meet all of the eligibility criteria for a cardiac resynchronization therapy device, patients who received a non–evidence-based ICD in our study were more likely to belong to a racial minority group other than black.¹⁴ Given that racial minorities have been shown to be less likely than white patients to receive evidence-based ICDs, it is concerning that some racial minority groups in this study were more likely to be recipients of a non–evidence-based device.¹⁵ Reasons for this finding need to be examined and addressed.

In our study, the rate of non–evidence-based ICD implants was significantly higher for nonelectrophysiologists than electrophysiologists. Potential reasons for this disparity include better knowledge of the data on primary prevention ICDs and increased commitment to adherence to practice guidelines by electrophysiologists. Future research should investigate actual reasons behind this disparity and propose ways to decrease non–evidence-based ICD implants.

There was no clear decrease in the overall number of non–evidence-based ICD implants over time. Although the proportion of non–evidence-based ICD implants decreased significantly from 2006 to 2007, there was no significant change from 2007 to 2009 to support a declining trend over time. These findings highlight the importance of continuing to enhance health care practioners' knowledge of practice guidelines. The only subgroup of patients who seemed to show a significant decrease in non–evidence-based ICD implants over time (as a proportion of all implants) was patients who received an ICD within 40 days of an MI. Although subgroup analyses should be viewed cautiously, potential reasons for this decline may include wider dissemination of evidence-based practice guidelines, participation in the ICD Registry, and other quality-improvement initiatives. Future accrual of data in the ICD Registry will allow a more robust examination of changes in non–evidence-based ICD implants over time.

Our study has several limitations. First, the analysis of complications was limited to events occurring in the hospital. However, in a previous study,¹⁶ our group demonstrated that most ICD-related complications occur in the hospital. Second, hospitals are required to submit data on Medicare patients only, so this analysis may not reflect all cases. However, 78% of the hospitals entered data on all patients undergoing ICD implantation.¹⁰ Our results likely reflect conservative estimates of non–evidence-based ICD implants because submitting data on primary prevention ICDs in Medicare patients is mandatory; and to receive payment, hospitals have incentive to ensure their procedures are performed for approved indications. It should be pointed out that some of these ICD implants may have been clinically appropriate. The guidelines clarify that “the ultimate judgment regarding care of a particular patient must be made by the physician and the patient in light of all of the circumstances presented by that patient. There are circumstances in which deviations from these guidelines are appropriate.”^{7 - 8}

In this study, we found that a substantial number of ICDs were implanted in patients who were similar to those who either were excluded from major clinical trials of primary prevention ICDs or shown not to benefit from ICD therapy in other trials. Such patients not only have more comorbidities than patients receiving an evidence-based device, but they are at a higher risk of in-hospital death and any postprocedure complication. We observed considerable variation in non–evidence-based ICD implants by site. The rate of non–evidence-based ICD implants was significantly higher for nonelectrophysiologists than electrophysiologists. There was no clear decrease in the overall number of non–evidence-based ICD implants over time. As such, more efforts should focus on enhancing adherence to evidence-based practice.