Design and rationale of the comprehensive evaluation of risk factors in older patients with AMI (SILVER-AMI) study

The recruitment network for SILVER-AMI was assembled with the goal of reflecting a wide geographic distribution and a diverse mix of practice types. There are approximately 70 recruitment sites located in the United States representing 27 states (Figure 2). The SILVER-AMI network is built upon the foundation of the recruitment network assembled for the VIRGO study [8] and additional sites were recruited through the American College of Cardiology’s Section of Geriatric Cardiology. The majority of sites (71.4%) are non-academic hospitals. Hospitals are heterogeneous in size (range: 83 to 2292 beds, median 475) and location (urban: 47.1%, suburban/rural: 52.9%).

Each site identifies a site research coordinator who has primary responsibility for implementing the study protocol. Prior to study initiation, the site research coordinators complete required trainings (in the recruitment of older hospitalized patients, obtaining of informed consent, and use of the electronic data capture system) conducted by the Yale Coordinating Center, and are supplied with standardized equipment to perform the physical assessments.

Screening is performed in the hospital setting by the site research coordinators who review daily admission records to identify potentially eligible participants. Site research coordinators then perform a medical record review to confirm eligibility in accordance with the criteria found in the Third Universal Definition of Myocardial Infarction [9], listed in Table 1.

If, during the screening process, the site research coordinator is concerned about the potential participant’s decisional capacity, the University of California, San Diego Brief Assessment of Capacity to Consent (UBACC) is administered [10]. If potential participants demonstrate impaired decision-making ability based on the UBACC, proxy consent is obtained in compliance with local regulations for legally authorized representatives.

In order to confirm study eligibility and provide further classification, medical records from enrolled participants are adjudicated by two physician investigators (JD, SC) at the Yale Coordinating Center, who confirm the baseline AMI diagnosis, interpret the diagnostic electrocardiogram (ECG), and determine whether the event represents ST elevation myocardial infarction (STEMI) or non ST elevation myocardial infarction (NSTEMI) based on previously published criteria [9].

Participants undergo an interview and physical assessment by the site research coordinator during their baseline AMI hospitalization (Table 2). In addition to sociodemographic and presentation characteristics, specific domains are tested which are described below. Participants receive a $25 gift card for completing this assessment.

To complement the participant interview and assessment, an in-depth medical record review is performed by the site research coordinator who collects information about clinical status at the time of the initial presentation (blood pressure, heart rate, presence of decompensated heart failure), comorbidities, laboratory results, in-hospital adverse events, and discharge disposition (Figure 3). Medical records are also provided to the Yale Coordinating Center where a research nurse (NL) conducts an in-depth review to collect information about medications, cardiac procedures, and discharge instructions.

The six-month telephone interview is conducted by staff at the Yale Coordinating Center. For participants with impaired decisional capacity at baseline, the six-month interview is completed with both the participant and assigned proxy.

The six-month interview repeats several key components of the baseline interview (Table 2), including questions about both general and disease-specific health status. In addition to the outcomes of re-admission and death, we define a decline in health status as a 10% reduction in overall quality of life as measured by the SF-12 [25]. Participants are also questioned on whether they have been re-hospitalized, and whether they have experienced any signs or symptoms that they attribute to medications (i.e., ‘adverse effects’) since their AMI discharge. If present, the participant is asked the medication to which they attribute the symptoms and if any action was taken to address the adverse effect. Participants receive an additional $50 gift card for completing the six-month interview.

At the six-month time point, the site research coordinator collects medical records on any hospital readmissions, outpatient cardiac procedures, emergency department (ED) visits, and deaths from the index hospital, as well as all other hospitals the participant reported using at the time of baseline interview. Medical records are provided to the Yale Coordinating Center where the events are reconciled with the participant’s self-reported hospitalizations during the 6-month phone interview. Any outstanding records are then collected. Medical records and death certificates are reviewed by physician investigators to determine whether the reported events represent true hospital admissions, as well as the primary discharge diagnosis, and/or cause of death.

The baseline interview is completed by the site research coordinator and transmitted electronically to the Yale Coordinating Center within three days (Figure 3). A Yale research assistant reviews the form and queries the site for missing or incomplete items. The site also transmits medical records, including emergency department and inpatient notes, ECGs, cardiac catheterization and operative (CABG) reports, and echocardiogram results, for abstraction by a research nurse at the Yale Coordinating Center.

All study data are managed at the Yale Coordinating Center using REDCap, an NIH-supported, HIPAA-compliant web-based Electronic Data Capture (EDC) system [26]. A custom Site Portal website manages the various administrative and scientific site workflows, including: site enrollment, IRB tracking; study document distribution, staff training, participant enrollment, invoicing, and site performance monitoring. In addition, we have integrated REDCap’s error detection and resolution features into a Data Query Quality Control system (DQCQ) that allows for error checks of varying complexity using SAS programs.

The primary outcomes of SILVER-AMI are the occurrence of all-cause hospital readmission, all-cause death, and decline in health status within six months of hospital discharge. Using independent risk factors that include cardio-centric factors (e.g. STEMI vs. NSTEMI, systolic blood pressure, Killip Class), as well as geriatric impairments, we will construct risk stratification models that predict the probability of each outcome for specific combinations of the risk factors. These models will be variants of the Cox regression models for the time-to-event outcomes (hospital readmission and mortality) and the logistic model used for decline in health status. Because mortality serves as a competing risk for hospital readmission, we will evaluate the associations of explanatory variables with both time-to-event outcomes using the competing risk approach of Fine and Gray [27]. All linearity assumptions will be checked graphically and deviations from model additivity will be accommodated by an examination of all statistically significant, clinically indicated two-factor interactions. All missing data will be examined for the nature of missingness, i.e., whether missing at random, and if justifiable, multiple imputation will be applied. Models will be evaluated with an examination of residuals and goodness-of-fit statistics using SAS/STAT - V9.3 (SAS Institute, Cary, NC) or later, with statistical significance defined as a two-sided p-value <0.05.

Internal and external validation techniques will then be used to test the reproducibility of the risk stratification tools. For internal validation, jackknife methods [28] will be used to evaluate the percent of correct classification; external validation will be performed on the 500 randomly selected participants who were excluded from both model building and internal validation.

Once the final models have been developed, user-interface applications will be developed for ‘smart-phones’ and other hand-held devices. The model coefficients for each outcome will be stored in hand-held applications whose outputs will provide the probability of each outcome. The inputs to the applications will consist of simple yes/no entries denoting the presence or absence of risk factors. Calculation will be presented in real time, and the results will list specific risk factors, and probability of the outcomes.

Of the SILVER-AMI cohort of 3000 participants, 2500 participants will serve as the development cohort and 500 will be randomly selected as the validation cohort. Based on 2008 Medicare data, we posited outcome rates of 39% and 28% for all-cause readmission and mortality. Based on previous research [29], we also assumed that a 10% decline in health status, as measured by the SF-36, would occur in 35% of surviving participants. With regard to the explanatory variables of primary interest, i.e., geriatric conditions and socio-demographic and clinical factors, we assumed each would exhibit measures of relative risk -1.3 for each time-to-event outcome and -1.5 for decline in health status. The 2500 participants in the development cohort provide 90% power to detect the stated magnitudes of relative risk for prevalence of the explanatory variables between 30% and 50%. For the validation sample, the accepted standard for time-to-event and logistic models should be large enough to contain a minimum of 100 outcome events [30]. For mortality, the least frequently expected outcome, the additional sample of 500 should allow for 140 events. The validation size of 500 will be preserved regardless of overall mortality and dropout.