Analysis of Total Events and Hospitalizations in the SPARCL Study

Lead Investigator: Michael Szarek, SUNY Downstate Medical Center
Title of Research Proposal: Analysis of Total Events and Hospitalizations in the SPARCL Study
Vivli Data Request: 4327
Funding Source: None
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Clinical benefit in endpoint-directed studies of cardiovascular (CV) disease is often characterized as an intervention reducing the risk of the first event in a composite of related non-fatal CV events and either all-cause or cause-specific death. By design, additional events in the composite that may occur after a patient’s first non-fatal event are available for analysis, and patients are typically encouraged to remain on randomized therapy through the first reported non-fatal event due, in part, to uncertainty the initial event will qualify as first in the composite. Consequently, the analysis of only the first event in the composite does not capture the entire impact of the interventional treatment on all documented events.

Several studies have demonstrated the benefits of intensive statin therapy on reducing first and subsequent non-fatal CV events in the settings of stable coronary heart disease (CHD) and post-acute coronary syndrome (ACS) (Tikkanen 2009; Murphy 2009; LaRosa 2010; Murphy 2016; Schwartz 2017), and similar findings have been reported with other interventional classes (White 2014). In these settings, follow-up for total non-fatal events is frequently interrupted by a competing event of death. Since it is anticipated that occurrences of non-fatal CV events will be associated with an increased risk of subsequent fatal events, deaths likely violate the non-informative censoring assumption for non-fatal events, regardless of whether all-cause or cause-specific deaths are included in the composite of total non-fatal and fatal events. Additionally, the statistical methods used in these prior analyses do not account for and/or quantify the possible association between total non-fatal and fatal events. Since these studies have not demonstrated consistent reductions in death associated with interventional therapy, it is not known if this could be explained by the nature of the association between non-fatal and fatal events within the study. For example, there may be a weak association between non-fatal and fatal events within these trials, and therefore reducing the risk of total non-fatal events would not necessarily be expected to modify the risk of fatal events.

To overcome these limitations, a novel approach is proposed to jointly model total non-fatal and fatal events in a pre-specified analysis of the SPARCL trial, allowing for the possibility that patients may experience multiple related non-fatal events. The method formally quantifies the association between non-fatal and fatal events while accounting for the fact that competing terminal events interrupt the non-fatal events process.

Note that no known analyses of total events in SPARCL have been completed to date.

Statistical Analysis Plan:


Hazard functions for total (first and subsequent) non-fatal CV events and all-cause death will be estimated by a joint semiparametric model (Rondeau 2012). A common frailty term is included in the two hazard functions, which can influence each differently due to an exponential association parameter. An association parameter of 0 indicates no association between a given patient’s risk of non-fatal CV events and death, whereas a value >0 indicates that patients who are at higher risk of non-fatal CV events are also at greater risk for death. Like Cox regression, treatment effects on non-fatal and fatal events will be separately summarized by hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). Point estimates and corresponding 95% CIs are also generated for the frailty variance and association parameter.


A sensitivity analysis will treat non-fatal cerebrovascular events (non-fatal stroke and transient ischemic attack [TIA], carotid revascularization) and non-fatal coronary events (myocardial infarction [MI, unstable angina, angina or ischemia requiring emergency hospitalization, coronary revascularization) as separate types of potentially recurrent events. This will result in the estimation of three hazard functions, where each non-fatal function contributes a shared frailty term to the fatal function.

Additional sensitivity analyses will include factors that are expected to be prognostic for survival (e.g., age, baseline Low-density lipoprotein cholesterol [LDL-C], diabetes status, etc.) to determine if inclusion of these factors in the hazard functions for non-fatal and fatal events modifies the relationship between the functions.


The joint models will be estimated with the frailty pack package in R (Rondeau 2012).

Requested Studies:

Data Contributor: Pfizer Inc.
Study ID: NCT00147602
Sponsor ID: A2581138

Public Disclosure:

Szarek M, Amarenco P, Callahan A, DeMicco D, Fayyad R, Goldstein LB, Laskey R, Sillesen H, Welch KM; SPARCL Committees and Investigators. Atorvastatin Reduces First and Subsequent Vascular Events Across Vascular Territories: The SPARCL Trial. J Am Coll Cardiol. 2020 May 5;75(17):2110-2118. doi: 10.1016/j.jacc.2020.03.015. Epub 2020 Mar 16
doi: 10.1016/j.jacc.2020.03.015