Cardiac Resynchronization Therapy

In comparing the survival distributions of two or more groups (for example, new therapy vs standard of care), Kaplan-Meier estimation¹ and the log-rank test² are the basic statistical methods of analyses. These are non-parametric methods in that no mathematical form of the survival distributions is assumed. If an investigator is interested in quantifying or investigating the effects of known covariates (e.g., age or race) or predictor variables (e.g., blood pressure), regression models are utilized. As in the conventional linear regression models, survival regression models allow for the quantification of the effect on survival of a set of predictors, the interaction of two predictors, or the effect of a new predictor above and beyond other covariates.
Among the available survival regression models, the Cox proportional hazards model developed by Sir David Cox³ has seen great use in epidemiological and medical studies, and the field of nuclear cardiology is no exception. What follows are some examples of Cox models being used in nuclear cardiology. Xu et al^{4 (link)} looked at how myocardial scarring (assessed with positron emission tomography [PET] or single photon emission computed tomography [SPECT]) and other demographic and medical history factors predicted mortality in patients with advanced heart failure who received cardiac resynchronization therapy. Bourque et al^{5 (link)} looked at how left ventricular ejection fraction (LVEF, assessed with angiography) and nuclear summed rest score (SRS, assessed with SPECT) interacted to change the risk of mortality. Hachamovitch and Berman^{6 (link)} looked at the incremental prognostic value of myocardial perfusion SPECT (MPS) parameters in the prediction of sudden cardiac death. Nakata et al^{7 (link)} looked at how the heart-to-mediastinum ratio (assessed with metaiodobenzylguanidine [MIBG] imaging) predicted cardiac death.
Survival models other than the Cox model have been used in nuclear cardiology as well. For example, in a study of diagnosis strategies for quantifying myocardial perfusion with SPECT, Duvall et al^{8 (link)} utilized a log-normal survival model, a member of the parametric family of regression survival models, since initial data exploration revealed that the proportional hazards assumption of the Cox model was invalid. While this is an excellent example of when to utilize other survival models, it has been more common to see such data presented in conjunction with a Cox model analysis. In earlier studies of MPS-derived predictors of cardiac events, Hachamovitch et al^{9 (link)} used Cox models to identify significant predictors and parametric models, specifically the accelerated failure time (AFT) model, to make estimates of the time to certain percentiles of survival. An identical analysis strategy was used by the research group comprised of Cuocolo, Acampa, Petretta, Daniele et al^{10 (link)–13 (link)} in their research of the impact of various SPECT-derived predictors on the occurrence of cardiac events.

We used VDW files to obtain data on comorbidities and procedures from inpatient and ambulatory healthcare encounters using ICD-9 codes, lab test results, as well as ambulatory pharmacy databases and site-specific diabetes mellitus and cancer registries.
Prevalent HF was defined by any hospitalization or ambulatory HF diagnosis during the 5 years before the index date. During the 5 years before cohort entry and throughout the follow-up period, we also assessed patient records for diagnoses of acute myocardial infarction, unstable angina, coronary artery revascularization, stroke or transient ischemic attack, cerebrovascular disease, other thromboembolism, atrial fibrillation or flutter, ventricular fibrillation or tachycardia, mitral or aortic valvular heart disease, peripheral arterial disease, rheumatic heart disease, receipt of a pacemaker, receipt of cardiac resynchronization therapy, receipt of an implantable cardioverter defibrillator, dyslipidemia, hypertension, diabetes mellitus, hospitalized bleed, diagnosed dementia, diagnosed depression, chronic lung disease, chronic liver disease, mechanical fall, and systemic cancer using relevant ICD-9 codes and current procedural terminology codes that have been previously described.^{21 (link)}At baseline and during the follow-up period, ambulatory systolic and diastolic blood pressure measurements were identified from VDW vital sign files, whereas information on serum low-density lipoprotien and high-density lipoprotein cholesterol measurements and blood hemoglobin levels were ascertained from site ambulatory lab databases.