Effect of Statin Therapy on Mortality in Older Adults Hospitalized With Coronary Artery Disease

A Propensity-Adjusted Analysis

Daniel P. Rothschild, MD; Eric Novak, MS; Michael W. Rich, MD

Disclosures

J Am Geriatr Soc. 2016;64(7):1475-1479. 

In This Article

Methods

Patients

Individuals aged 80 and older consecutively discharged from Barnes-Jewish Hospital in St. Louis, Missouri, from January 2006 through December 2010 with a primary diagnosis of stable CAD, unstable angina pectoris, or AMI based on International Classification of Diseases, Ninth Revision, codes were identified retrospectively using the hospital's Clinical Investigation Data Exploration Registry database, which is derived from comprehensive electronic medical records. Individuals who died during the index hospitalization were excluded. For subjects with more than one qualifying admission during the study period, only the first admission was included. There were no other exclusion criteria. The Washington University Human Research Protection Office reviewed and approved the study, and a waiver for obtaining informed consent was granted because of the retrospective nature of the study, lack of participant contact, and the high mortality of study subjects during follow-up.

Data Collection

One thousand two hundred sixty-two individuals were identified who fulfilled the inclusion criteria. Data collection included demographic information (age, sex, race, marital status), basic laboratory tests (hemoglobin, creatinine, serum urea nitrogen, lipid profile), comorbidities (hypertension, diabetes mellitus, tobacco use, peripheral arterial disease, atrial fibrillation, heart failure, dementia), and cardiovascular medications prescribed at discharge. The Social Security Death Index was used to assess vital status for all subjects through June 1, 2014.

Statistical Analysis

Based on the presence of a discharge prescription for any statin, participants were divided into a statin group (n = 913) and a non-statin group (n = 349). The primary endpoint was all-cause mortality. Survival days were calculated from the date of discharge from the index admission until the date of death or June 1, 2014, whichever came first. Prespecified subgroup analysis according to aged 80 to 84 versus aged 85 and older was performed. Baseline characteristics of the statin and nonstatin groups were compared using Student t-tests, Fisher exact tests, and Kruskal-Wallis tests as appropriate. Kaplan-Meier curves were constructed to compare unadjusted survival in the statin and nonstatin groups. Cox proportional hazards models were created to examine the effects of age, statin therapy, and their interaction on all-cause mortality. Data imputation using sequential regression methods was used to estimate missing values for all variables for which the proportion of missing data did not exceed 10%.[7] Multiple imputation techniques were used, and five imputed data sets were created. Propensity scores for likelihood of receiving statin therapy were created using logistic regression models that incorporated all available variables. Two distinct propensity analyses were performed. In the first, a greedy-matching algorithm was used to create a 1:1 matched sample of participants who did and did not receive statins.[8] The number of matched pairs was different for each imputed data set, but on average, 81% of participants who did not receive statins were successfully matched. For the second analysis, inverse probability of treatment weighting (IPTW) was used to examine group differences. For both methods, standardized differences were calculated to examine covariate balance before and after propensity adjustment, with standardized differences of 10% or greater considered clinically meaningful. Cox models to examine survival among propensity score–matched subjects were built based on previously developed methods.[9] Models created under IPTW used robust standard errors developed using marginal models to account for weighting. Kaplan-Meier curves were created after IPTW adjustment. For all analyses, P < .05 was considered significant. SAS version 9.4 (SAS Institute, Inc., Cary, NC) was used for all main analyses. The SAS-callable software IVEware (University of Michigan) was used for data imputation.

processing....