Methods

Beginning in 2005, the Centers for Medicare and Medicaid Services required hospitals to submit patient-level data to an ICD registry as part of policy changes that provided Medicare coverage of the costs associated with implantation of the devices. Medicare beneficiaries represent nearly the entire population of older patients in the United States with primary prevention ICDs. Medicare registry records for all patients who underwent ICD implantation occurring from January 2005 to April 2006 were obtained for this study.

Medicare hospital claims files, program eligibility records, and dates of death were obtained for patients with records in the ICD registry, for the period from January 2004 through June 2010. Registry records were matched to Medicare eligibility and utilization data using patients’ Social Security numbers by the Centers for Medicare and Medicaid Services. Medicare hospital claims data included the dates of hospital admission and International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes that identified the primary cause of hospitalization and concurrent secondary conditions. Dates of death were obtained for deceased patients from the Medicare program eligibility file records.

Subpopulations of Medicare patients with records in the ICD registry were defined for patients meeting class I or class IIa recommendations for CRT for patients in sinus rhythm on the basis of the American College of Cardiology Foundation, American Heart Association, and Heart Rhythm Society 2012 device-based therapy guidelines that summarize the available evidence for therapeutic efficacy. Patients meeting class I criteria for CRT included those with left ventricular ejection fractions (LVEFs) ≤35%; those in NYHA class II, III, or IV; and those with left bundle branch block with QRS durations ≥150 ms. Class IIa criteria identify patients for whom it is reasonable to perform the procedure. Patients meeting class IIa criteria for CRT included those with LVEFs ≤35%; those in NYHA class II, III, or IV; and those with either left bundle branch block with QRS duration 120 to 149 ms or without left bundle branch block with QRS duration ≥150 ms. We did not include patients meeting the class IIa criteria of 40% right ventricular pacing or atrial fibrillation with atrioventricular node ablation or adequate pharmacologic rate control, because neither the percentage of ventricular pacing or adequacy of rate control could be assessed using the data available from the ICD registry.

Patients were followed for the occurrence of death from any cause and for the first occurrence of rehospitalization caused by heart failure (HF), for a minimum of 50 months and for up to 66 months, depending on the date of ICD implantation. The first occurrence of HF hospitalization was identified as the first inpatient hospital admission after ICD implantation with a primary International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis code of 428.0 to 428.9 (congestive HF). The ICD registry uses Internet-based data collection that limits the potential for missing values in reported data elements. Each of the categorical and continuous patient characteristics included in the analysis had known values.

Multivariate logistic regression was used to estimate the probability of receiving CRT-D among all patients included in the 2 ICD study populations. Separate models were estimated for the class I and class IIa groups. The 2 statistical models included a series of predictive covariates that were considered likely indicators for CRT-D or that were considered likely to be associated with either mortality or risk for rehospitalization caused by HF. All covariates were selected a priori on the basis of their clinical relevance, and all selected covariates were retained in the final models.

The multivariate logistic regression model’s capacity to discriminate between patients with and without CRT-D was measured using the C-statistic. A C-statistic of 0.5 indicates that the model provides no predictive discrimination, while a value of 1.0 indicates perfect discrimination between patients with CRT-D and those with ICDs only. The proportion of the total variation in the probability of CRT-D accounted for by the model was assessed using Nagelkerke’s R ² statistic. Values of Nagelkerke’s R ² statistic range from 0 for models that provide no predictive information to 1 for models that predict perfectly. The Wald chi-square test statistic was used to measure the relative contribution of each individual covariate included in the model toward the overall performance of the model.

Every patient with CRT-D was matched to a patient in the group with standard ICDs (ICD only) with the closest propensity score (probability of receiving CRT-D). Some patients with ICD only were matched more than once to patients in the CRT-D patient group. All covariates included in the propensity score model were assessed for their distributional balance between the CRT-D and matched ICD-only groups. Differences in the distribution of propensity score covariate values between patients in the CRT-D and ICD-only groups were assessed for statistical significance using bivariate logistic regression analysis to calculate the Wald chi-square test statistic for the association between group status and covariate values.

Cox proportional-hazards regression analysis was used to estimate differences in mortality risk up to 4 years after implantation surgery, between patients with CRT-D and those with ICDs only. Survival functions for the CRT-D patient group and for the matched cohort of patients from the ICD-only group were estimated with stratification on the matched pairs, to account for the matched nature of the study data.

Competing risks analysis was used to estimate differences in the risk for rehospitalization caused by HF up to 4 years after implantation surgery. Because some patients may die without previous HF hospitalization, analysis of HF hospitalization events after implantation requires adjustments for differences in the occurrence of death as a competing event. Competing risks analysis using the semiparametric proportional-hazards model was used to compare CRT-D and ICD-only patients with regard to their risk for HF hospitalization.

The sensitivity of the results to a potentially missing confounder was assessed by repeating the complete analysis using a revised propensity score model that excluded a highly significant covariate in the original model. We also assessed the sensitivity of the results to the inclusion of additional adjustments for patient differences in the propensity score–balanced groups. Some studies using propensity scores include the individual covariates used to derive the propensity scores as additional adjustment covariates in the comparison of patient outcomes. The original comparison of survival in the matched population using Cox proportional-hazards regression analysis was repeated, including each of the covariates originally used to estimate the probability of receiving CRT-D as additional covariates. The competing risks analysis was also repeated, including a subset of covariates expected to contribute the most to the statistical performance of the model.

Results

The ICD registry included 6,496 patients with CRT-D and 2,079 patients with ICDs only who met class I recommendations for CRT. The multivariate logistic regression model used to calculate the probability of CRT-D for these patients achieved adequate statistical performance, obtaining a C-statistic of 0.76 and a Nagelkerke’s R ² value of 0.25. Table 1 lists each of the 35 covariates included in the propensity score model, along with the Wald chi-square test statistic values and associated p values for each covariate. The Wald chi-square test statistic values demonstrate that the covariates NYHA class, implantation personnel type, facility type, history of ventricular arrhythmias, use of diuretics, and use of angiotensin receptor blockers were leading individual contributors to the predictive performance of the model. Lists of the specific value distributions for each covariate are available from the authors in a Supplementary Table .

The distributions of covariate values between patients in the CRT-D and matched ICD-only groups were balanced for most of the covariates included in the propensity score model. Statistically significant differences (p <0.01) occurred for 10 of the 35 covariates included in the propensity score model. Only 2 of these covariates were significant predictors of CRT-D: chronic kidney disease and the use of angiotensin receptor blocker. Patients with ≤8 matches to patients in the CRT-D group accounted for 54% of the patients in the ICD-only group. The absolute difference in propensity scores between matched pairs was <0.005 for 99.8% of the matched population.

Table 2 presents hazard ratios (HRs) for the difference in the risk for death with CRT-D compared with that for patients in the ICD-only group, for patients who met class I recommendations. The relative hazard of death was significantly lower for patients with CRT-D (HR 0.83). These results were not sensitive to the potential effects of a missing covariate in the propensity score model. Nearly equivalent results were obtained by repeating the complete analysis, with adjustments for history of ventricular arrhythmia (Wald chi-square 27.9, p <0.0001) excluded from the propensity score model (HR 0.82). Repeating the analysis with supplemental adjustments for the effects of all covariates included in the propensity score model yielded results similar to the original comparison (HR 0.87). Figure 1 illustrates the cumulative incidence of death over the 4 years of available follow-up after implantation for patients with CRT-D or ICDs only in the class I study population.

Table 2

Risk of all cause death or heart failure hospitalization in the CRT-D registry group compared to the propensity score matched ICD only group for class I patients

	Hazard Ratio	95% C.I.	p-Value
All cause death over 4 years of follow-up
CRT-D compared to matched ICD group	0.83	(0.77–0.88)	<0.0001
CRT-D compared to matched ICD group, estimated using original propensity score model with excluded covariate	0.82	(0.77–0.88)	<0.0001
CRT-D compared to matched ICD group, estimated with concurrent adjustments for covariates used in propensity score model	0.87	(0.81–0.94)	0.0002
Hospitalization for heart failure over 4 years of follow-up
CRT-D compared to matched ICD group	0.88	(0.83–0.94)	<0.0001
CRT-D compared to matched ICD group, estimated using original propensity score model with excluded covariate	0.82	(0.77–0.87)	<0.0001
CRT-D compared to matched ICD group, estimated with concurrent adjustments for covariates used in propensity score model	0.88	(0.83–0.93)	<0.0001

Only gold members can continue reading. Log In or Register to continue

Share this:

• Click to share on Twitter (Opens in new window)
• Click to share on Facebook (Opens in new window)

Related

Related posts:

1. Coronary Atheroma Regression and Plaque Characteristics Assessed by Grayscale and Radiofrequency Intravascular Ultrasound After Aerobic Exercise
2. Comparing Changes in Severe Versus Mild Perfusion Defect Size in Patients Who Underwent Serial Rubidium-82 Positron Emission Tomography Myocardial Perfusion Imaging
3. Aldosterone-Cortisol Imbalance Immediately After Fontan Operation With Implications for Abnormal Fluid Homeostasis
4. Gluten Sensitivity: New Epidemic or New Myth? Every Major Change in Our Diet Carries With It the Possibility of Unforeseen Risks

Stay updated, free articles. Join our Telegram channel

Join