ARTEMIS

Affordability and Real-world Antiplatelet Treatment Effectiveness After Myocardial Infarction Study

Statistical Analysis Plan

D5130R00030
Version date: April 3, 2017
Approved: November 30, 2017
Prepared by: DCRI ARTEMIS Team
[Confidential]

ARTEMIS

Affordability and Real-world Antiplatelet Treatment Effectiveness After Myocardial Infarction Study
Study Statistician
Statistical Analysis Plan
Statistical Analysis Plan

Table of Contents

  1. Overview ..... 6
    1.1 Primary Objectives ..... 6
    1.2 Secondary Objectives ..... 6
    1.3 Site and Patient Inclusion Criteria ..... 6
    1.4 Sample Size Justification ..... 7
  2. General Considerations for Data Analysis ..... 9
    2.1 Analysis Datasets and Baseline Comparisons ..... 9
    2.2 Analysis Populations ..... 10
    2.3 Accounting for Hospital Clustering: Marginal vs. Conditional Modelling Approaches for Binary Endpoints ..... 10
    2.4 Missing Data ..... 10
    3 Primary Endpoints ..... 10
    3.1 Non-persistence with antiplatelet therapy at 1 year ..... 10
    3.2 MACE ..... 11
  3. Secondary Endpoints ..... 12
    4.1 Medication Selection at discharge ..... 12
    4.2 Non-persistence with antiplatelet therapy at 1 year ..... 12
    4.3 Medication fill using pharmacy data only ..... 12
    4.4 Medication Drug Levels ..... 13
    4.5 Comparison of Medication Use from different sources ..... 13
    4.6 Cost ..... 13
    4.7 Safety - Bleeding ..... 13
    4.8 Unplanned Revascularization ..... 13
    4.9 Components of MACE ..... 13
    4.10 MACE plus unplanned revascularization ..... 14
    4.11 Cardiovascular Mortality ..... 14
    4.12 As Treated Analysis ..... 14
    4.13 Exploratory Analysis ..... 14
  4. Subgroups ..... 14
  5. Secondary Analysis of Primary Endpoints ..... 14
    6.1 Non-persistence with antiplatelet therapy at 1 year ..... 14
    6.2 MACE ..... 15
    Statistical Analysis Plan
  6. References ..... 16
  7. Appendices ..... 17
    8.1 Appendix: Table of Contents for Statistical Tables, Figures, and Listings ..... 17
    8.2 Appendix: Table Shells ..... 18
    8.2.1 Main Table Shells ..... 18
    8.2.2 Supplementary Table Shells ..... 30
    8.3 Appendix: List of Adjustment Variables ..... 35
    8.3.1 Adjustment Variables for Primary Models ..... 35
    8.3.2 Variables for Propensity Score Model ..... 36
    8.4 Appendix: Figure Shells ..... 38
Statistical Analysis Plan

1. Overview

The purpose of the statistical analysis plan is to describe the key components of the Affordability and Real-world Antiplatelet Treatment Effectiveness After Myocardial Infarction Study final data analysis. This plan is a supplement to the materials provided in the ARTEMIS protocol (version date: March 12, 2015).
ARTEMIS is a prospective cluster-randomized clinical trial that will evaluate whether patient copayment reduction significantly influences antiplatelet selection and long-term adherence. This study will also examine patient outcomes and the overall cost of care after AMI. After IRB approval, sites will be randomized to either the intervention or the control. Randomization will be stratified by annual site AMI volume and proportion of ticagrelor use using medians across potential sites as cut-offs for identifying high vs. low categories. Approximately 11,000 patients with STEMI or NSTEMI will be enrolled at approximately 300 hospitals.

1.1 Primary Objectives

The co-primary objectives of ARTEMIS are the following:
To determine if patient copayment reduction leads to higher long-term persistence of any receptor inhibitor at one year after discharge.
To determine if patient copayment reduction leads to lower risk of MACE (composite of death, AMI, and stroke) at one year after discharge.

1.2 Secondary Objectives

To evaluate whether reducing patient copayments affects selection of receptor inhibitor medication at discharge.
To assess the impact of copayment reduction on the total cost of health care for patients after AMI.

1.3 Site and Patient Inclusion Criteria

Study Site Selection Criteria

Hospitals are eligible to be included in the study if they meet the following criteria:
[1] Treat at least 50 STEMI or NSTEMI patients annually
[2] Have clopidogrel and ticagrelor available for clinical use on their hospital formulary

Study Patient Selection Criteria

Patients are eligible to be included in the study if they meet all of the following criteria:
[1] Are years of age
Statistical Analysis Plan
[2] Have been diagnosed with STEMI or NSTEMI during the index hospitalization
  • STEMI is defined as symptoms of cardiac ischemia (e.g., chest pain) associated with either a new left bundle branch block or ST-segment elevation of in at least two contiguous leads on the electrocardiogram (ECG). If no reperfusion treatment is pursued, patients must be treated with primary PCI or fibrinolytic therapy, or have at least one troponin I, troponin T, or creatine kinase-MB value greater than the institutional upper limit of normal.
  • NSTEMI is defined as symptoms of cardiac ischemia associated with a rise and fall in biomarkers indicating myocardial necrosis. At least one troponin I, troponin T, or creatine kinase-MB value must be greater than the institutional upper limit of normal.
    [3] Are treated with a receptor inhibitor at the time of enrollment
    [4] Have United States-based health insurance coverage with prescription drug benefit
    [5] Have been fully informed and are able to provide written consent for longitudinal follow-up
Patients are excluded if they meet any of the following criteria:
[1] Have a history of prior intracranial hemorrhage
[2] Have any contraindications to P2Y receptor inhibitor therapy at discharge
[3] Involvement in another research study that specifies the type and duration of receptor inhibitor use within the next 12 months
[4] Have a life expectancy of less than one year
[5] Have plans to move outside the United States in the next year

1.4 Sample Size Justification

The proposed sample size has been determined to provide adequate statistical power for the coprimary study objectives related to the copayment reduction intervention. The co-primary objective determines whether patient copayment reduction leads to greater persistence to receptor inhibitor therapy at one year after hospital discharge. The hypothesis underlying this objective is that reducing patient copayment in a contemporary population of AMI patients will result in a significant increase in persistence to receptor inhibitor therapy at one year, when compared with usual care. An increase of in the persistence to receptor inhibitor therapy would be considered a clinically important difference[1]. To achieve this objective with a patient-level randomization design, a sample size of 5392 patients would provide greater than power with a two-sided Type I error rate of 0.05 . A sample size of 4622 patients would provide greater than power under the same assumptions. These power calculations are based on the assumption that the expected one-year persistence rate in the control group is . These calculations are based on the two group continuity corrected chi-square test statistic and assume that all observations are independently distributed. However, the sample size needs to be adjusted due to the cluster randomized design. We have applied the method described by Eldridge et al. [2] which accounts for the coefficient of variation (CV) of cluster

Statistical Analysis Plan

size and the intra-cluster correlation (ICC). Based on prior multicenter studies, we anticipate an ICC for this endpoint of approximately 0.025 . Assuming a total of 300 sites randomized with an average sample size of 36.67 patients per site and a CV of 0.65 would yield a design effect of approximately 2.28. The CV of 0.65 has been suggested by others and can be guided by providing minimum and maximum enrollment at the site level [2]. Therefore, a total sample size of 11,000 patients enrolled at 300 sites would result in an effective sample size of 4827 and be sufficient to provide between and power to detect an absolute difference between treatment groups in the cluster randomized design.
For the one-year MACE endpoint, the underlying hypothesis is that patient copayment reduction leads to a reduction in MACE risk; this is partially due to the selection of a more potent antiplatelet agent that has been shown to reduce MACE risk in randomized clinical trials, and partially due to greater persistence of an evidence-based secondary prevention medication. For this endpoint, we have assumed a control group event rate of . A clinically meaningful event reduction of would yield a one-year event rate of [3]. To achieve power with a patient-level randomization, a 1:1 allocation ratio, and a two-sided Type I error rate of 0.05 would require a total of 6728 patients. Under the same assumptions, a total sample size of 7670 would provide power. These sample size estimates are based on the continuity corrected chi-square test. Since the unit of randomization will be the site rather than the individual, we again need to consider the correlation of response within site and the CV on the number of patients enrolled per site. Prior multicenter studies have suggested an ICC of approximately 0.01 for the MACE endpoint [2]. A total sample size of 11,000 patients enrolled at 300 sites, assuming an ICC of 0.01 and a CV of 0.65 , would yield an effective sample size of 7278 patients (Table). Therefore, the total sample size of 11,000 patients enrolled at 300 sites would be expected to provide between and power to detect an relative reduction in MACE ( vs. ).
Table: Required Sample Size for the MACE Endpoint
Sites Total Sample Size Average # of Patients per Site ICC CV Effective Sample Size
250 10000 40 0.010 0.65 6414
250 11000 44 0.010 0.65 6808
250 12000 48 0.010 0.65 7174
250 13000 52 0.010 0.65 7516
250 14000 56 0.010 0.65 7836
250 15000 60 0.010 0.65 8137
250 16000 64 0.010 0.65 8420
250 17000 68 0.010 0.65 8686
250 18000 72 0.010 0.65 8936
300 10000 33.33 0.010 0.65 6830
300 11000 36.67 0.010 0.65 7278
300 12000 40 0.010 0.65 7698
300 13000 43.33 0.010 0.65 8092
300 14000 46.67 0.010 0.65 8466
300 15000 50 0.010 0.65 8818
300 16000 53.33 0.010 0.65 9150
Statistical Analysis Plan
300 17000 56.67 0.010 0.65 9465
300 18000 60 0.010 0.65 9764
Patient-reported persistence to antiplatelet therapy will be validated using pharmacy records that will be collected on a subset of the overall study population. The assumed standard deviation of 0.25 (for the proportion of days covered) is based on a recent randomized clinical trial to assess an intervention designed to improve adherence [4]. Assuming 1:1 randomization and a two-sided Type I error rate of 0.05 , a sample size calculation based on the two-sample t-test suggests that a total sample size of 1644 patients will yield and a total sample size of 2034 will yield power. A random sample of 2400 patients from the 300 sites randomized with a CV of 0.25 and an ICC of 0.025 would yield a design effect of approximately 1.26 . The resulting effective sample size of 2021 patients would yield approximately power to detect a difference of between the patient copayment reduction intervention and control groups. The target sample size of 2500 patients with pharmacy records allows for missing data due to records that are not available.

2. General Considerations for Data Analysis

We will include a detailed flow diagram showing the number of participants' eligible, number of subjects randomized to Copayment Invention and Usual Care arms, numbers of subjects lost to follow-up or excluded from analyses, and the number of subjects analyzed for the key study endpoints. Additionally, we will describe the number of clinical sites in Copayment Intervention and Usual Care arms including the mean (median, range) for key study elements by site. Key elements will include the following items: descriptions of participants, interventions, objectives, outcomes, and sample size justification; and details about the randomization procedure, factors used to stratify randomization, (lack of) blinding, statistical methods, participant flow, dates of recruitment, baseline data by individuals and by cluster, numbers analyzed, outcomes and estimation, adverse events, and a discussion. These elements are based on the CONSORT statement for cluster randomized studies [5].
After Steering Committee consensus, the randomization scheme was changed from to Usual Care vs. Intervention effective November 16, 2015. Randomization schema was added below as a covariate for adjustment.
All analyses will be conducted using SAS version 9.4 or higher software. All tests will be twosided and a p-value of will be considered statistically significant.

2.1 Analysis Datasets and Baseline Comparisons

Data from all enrolled patients, regardless of whether or not they completed all protocol followup requirements, will be included for analysis. Baseline comparisons of patient characteristics and randomization stratification variables between intervention and control arms groups will be summarized as the mean; standard deviation; median; and percentiles for continuous variables; and as counts and percentages for categorical variables. We will present baseline
Statistical Analysis Plan
characteristics and randomization stratification variables at both the patient and cluster levels. Cluster level summary data will be presented as means (standard deviations). All study objectives will be analyzed using intention-to-treat analyses.

2.2 Analysis Populations

The study population for primary and secondary analyses in ARTEMIS will start with all enrolled patients who survived the index MI hospitalization and did not withdraw from the study before hospital discharge. Since the intervention arm voucher provides copayment assistance for a generic (clopidogrel) or a brand (ticagrelor) receptor inhibitor, we will conduct the primary endpoint analyses first among patients discharged on either clopidogrel or ticagrelor, and then repeat the analysis among all patients regardless of discharge receptor inhibitor type.

2.3 Accounting for Hospital Clustering: Marginal vs. Conditional Modelling Approaches for Binary Endpoints

There are two general approaches that account for within hospital correlation in a statistical model; they are 1) marginal or population-averaged model and 2) conditional or subject-specific model. These two approaches differ in interpretation of model estimates and the way that correlation of measurements are incorporated in the model. For example, under the marginal model, the exponentiated treatment coefficient represents the odds of an average patient in the treatment group to be persistent compared to an average patient in the control group. Under the conditional random-effects model, the exponentiated treatment coefficient represents the odds of persistence for a treated person compared to the same person if they were not treated. As stated in the protocol we will use population averaged methods as our primary approach to account for hospital clustering.

2.4 Missing Data

Operational efforts will be made to minimize missing data at baseline and during follow-up. During the enrollment phase, baseline data will be reviewed on a monthly basis and sites will be notified regarding any data quality concerns. Study personnel will confirm that missing data cannot be obtained. Follow-up data will be regularly reviewed after 50% of patients reach 3 months post-discharge.
We will impute socioeconomic variables, lab values, and weight to age, gender, and race specific modes for categorical variables and medians for continuous variables. Medical history, home medications, admission features, and in-hospital events will be imputed to the mode.

3 Primary Endpoints

3.1 Non-persistence with antiplatelet therapy at 1 year

The co-primary endpoint of long-term non-persistence will be assessed using patient-reported medication non-persistence. Permanent and temporary discontinuation of a receptor inhibitor will be queried at each follow-up interview. For patients with missing patient-reported medication information, pharmacy fill data will be used to ascertain persistence. Patients who have continued receptor inhibitor use at one year from discharge with less than 30
Statistical Analysis Plan
continuous days of interruption will be considered persistent. We will use the last observation carried forward (LOCF) method for patients who died before one year or had missing 1 year status.
The co-primary objective determines whether patient copayment reduction leads to higher longterm persistence with antiplatelet therapy at one year. The study endpoint for this objective is the proportion of patients at one year who had an interruption days of receptor inhibitor. The primary analysis will be a logistic regression model with parameters estimated using generalized estimating equations (GEE) to account for within hospital clustering and adjustment for selected patient characteristics (Section 8.3.1) which includes a propensity score for intervention (Section 8.3.2). The propensity score will be estimated using a logistic regression model for intervention group. Categorical variables will be included as sets of indicator variables. Continuous variables will be included assuming a simple linear relationship. Balance of covariates between intervention and usual care arms will be assessed using standardized differences as recommended by Austin [6]. We will assess functional form and possible transformations of the propensity score to be included in the outcome model. Cluster heterogeneity will be quantified using ICCs calculated from unadjusted and adjusted models. We will calculate the ICCs across all hospitals and also by treatment group.
After Steering Committee consensus, the randomization scheme was changed from to Usual Care vs. Intervention effective November 16, 2015. Randomization schema was added as a covariate for adjustment as shown in Section 8.3.1 and to the propensity model as shown in Section 8.3.2.

3.2 MACE

The primary study objective evaluates whether patient copayment reduction leads to lower risk of MACE at one year. MACE is defined as the composite of all-cause death, recurrent myocardial infarction, and stroke. Follow-up will be censored at time of study withdrawal or last known alive. The time-to-first MACE event up to one year post-discharge will be compared between intervention and control arms. Cumulative incidence rates will presented as KaplanMeier curves and in tables for 30 days, 6 months, and 1 year post-discharge. The primary analysis will be a Cox proportional hazards model accounting for within hospital clustering using robust standard errors and adjustment for selected patient characteristics (Section 8.3.1) which includes a propensity score for intervention. We will use the same propensity score described in section 3.1. Cluster heterogeneity will be quantified using ICCs calculated from unadjusted and adjusted models. We will calculate the ICCs across all hospitals and also by treatment group.
After Steering Committee consensus, the randomization scheme was changed from 1:1 to 2:1 Usual Care vs. Intervention effective November 16, 2015. Randomization schema was added as a covariate for adjustment as shown in Section 8.3.1 and to the propensity model as shown in Section 8.3.2.
Statistical Analysis Plan

4. Secondary Endpoints

4.1 Medication Selection at discharge

Inclusion Criteria: All patients.
We will use logistic regression with GEE to account for within hospital clustering to evaluate whether copayment intervention is associated with discharge inhibitor type. We will use the same methods for adjustment as in the primary analysis.

4.2 Non-persistence with antiplatelet therapy at 1 year

Non-persistence or death vs. persistence at 1 year

Analogous to Section 3.1, we will examine the outcome of death or non-persistence; nonpersistence still defined as_interruption days of receptor inhibitor. The primary analysis will be a logistic regression model with generalized estimating equations (GEE) to account for within hospital clustering and adjustment for selected patient characteristics (Section 8.3.1) and a propensity score for intervention (Section 8.3.2). We will use the same propensity score described in section 3.1. Cluster heterogeneity will be quantified using ICCs calculated from unadjusted and adjusted models. We will calculate the ICCs across all hospitals and also by treatment group.

Non-persistence to initial receptor inhibitor at year

Analogous to Section 3.1, we will use the LOCF method for patients who died before one year or had missing 1 year P2Y inhibitor status. Patients who have continued their initial P2Y receptor inhibitor use at one year from discharge with less than 30 continuous days of interruption will be considered persistent. Patients who switched from their initial P2Y receptor inhibitor will be considered non-persistent.
The primary analysis will be a logistic regression model with generalized estimating equations (GEE) to account for within hospital clustering and adjustment for selected patient characteristics (Section 8.3.1) and a propensity score for intervention (Section 8.3.2). We will use the same propensity score described in section 3.1. Cluster heterogeneity will be quantified using ICCs calculated from unadjusted and adjusted models. We will calculate the ICCs across all hospitals and also by treatment group.

4.3 Medication fill using pharmacy data only

Inclusion Criteria: Patients with pharmacy data collected.

Persistence:

Using pharmacy data, we will calculate persistence using the same definition as in the primary analysis with non-persistence defined as a fill gap days. Analogous to Section 3.1, we will use the LOCF method for patients who died before one year or had missing 1 year P2Y status. Persistence at 1 year will be compared by copayment intervention using logistic regression with GEE to account for within hospital clustering and the same methods for adjustment as in the primary analysis.
Statistical Analysis Plan

Adherence:

Using pharmacy data, we will calculate the proportion of days covered. Patients with proportion of days covered of expected prescriptions over one year of follow-up or until death date will be considered adherent. Adherence at 1 year will be compared by copayment intervention using logistic regression with GEE to account for within hospital clustering and the same methods for adjustment as in the primary analysis.

4.4 Medication Drug Levels

Inclusion Criteria: All patients with valid drug level data.
A subset of patients will have blood drawn over the one year of follow-up after AMI. This blood draw will be randomly assigned to 250 patients ( 125 in each arm) at each of the following follow-up time points: , or 12 months. Drug levels or metabolites of clopidogrel or ticagrelor will be measured, as appropriate. Persistence (yes vs. no) will be defined based on clinically selected cut-offs of minimum drug levels. Medication drug level data will be presented in Table 6 in Section 8.2.1.

4.5 Comparison of Medication Use from different sources

Overall and stratified by copayment reduction, we will assess agreement among patient reported medication use, medication fill adherence, and drug levels using Kappa statistics and summarize using frequencies and percentages (Table 7 in Section 8.2.1).

4.6 Cost

A detailed description of the healthcare resource utilization endpoints and analysis will be contained in a separate SAP, drafted following and guided by the primary clinical endpoint results.

4.7 Safety - Bleeding

Bleeding events will be collected using the Bleeding Academic Research Consortium (BARC) bleeding definition. Additionally, the severity of bleeding will be categorized using the Global Utilization of Streptokinase and t-PA for Occluded Coronary Arteries (GUSTO) definition for severe, moderate, or mild bleeding. We will assess the effect of copayment reduction on bleeding using the same methods as the composite MACE outcome.

4.8 Unplanned Revascularization

We will assess the effect of copayment reduction on unplanned revascularization using the same methods as the composite MACE outcome.

4.9 Components of MACE

We will assess the effect of copayment reduction on each component of MACE using the same methods as the composite MACE outcome.
Statistical Analysis Plan

4.10 MACE plus unplanned revascularization

We will assess the effect of copayment reduction on the composite of all-cause death, recurrent MI, stroke, and unplanned revascularization using the same methods as the composite MACE outcome.

4.11 Cardiovascular Mortality

We will assess the effect of copayment reduction on cardiovascular mortality using the same methods as the composite MACE outcome.

4.12 As Treated Analysis

We will conduct an as treated analysis of the co-primary endpoints by excluding intervention patients who did not use the voucher during any of the 12 months of follow-up. We will use the same methods as described in Section 3.1 and 3.2, but we will re-fit the propensity score among this population. All tables included in Section 8.2.1 with the exception of Table 1 will be repeated for this subset of patients.

4.13 Exploratory Analysis

Reduction of the co-primary endpoint of MACE could be driven by several factors including improved adherence to therapy, increasing use of a higher-potency receptor inhibitor, or both. Exploratory analyses will examine the associations of these two factors with MACE. In addition, instrumental variable analyses will be considered for the co-primary endpoints.

5. Subgroups

We will conduct subgroup analyses (Tables 3b, 4b, and 9b) for the following subgroups using the same methods as the primary analysis with the addition of the main effect for subgroup and an interaction term for subgroup by intervention. Indicator variables for the subgroups will be included in the propensity model for all subgroups except initial treatment selection.
  • Age: Age and age <65
  • Sex: Males and females
  • Insurance status: private and non-private
  • Race: White and non-White
  • STEMI and NSTEMI
  • In-hospital PCI and no in-hospital PCI

6. Secondary Analysis of Primary Endpoints

6.1 Non-persistence with antiplatelet therapy at 1 year

Propensity matched analysis: Within each of the four randomization strata defined, using medians across potential sites from survey conducted prior to ARTEMIS, by high ( ) vs. low (<400) annual site AMI volume and high ( ) vs. low ( ) proportion of ticagrelor use and by randomization scheme ( vs. 1:1), we will match intervention arm patients to usual care patients. Matching within randomization scheme and strata forces matches across sites of similar size and proportion ticagrelor use and time of site randomization. To run the computerized

Statistical Analysis Plan

matching of intervention to usual care patients, we will utilize the gmatch macro publicly available from the Mayo Clinic Division of Biomedical Statistics and Informatics website of locally written SAS macros. This macro was downloaded on 5/13/13 from the following website: http://mayoresearch.mayo.edu/mayo/research/biostat/sasmacros.cfm. The gmatch macro performs greedy matching of cases to controls (intervention patients to usual care patients) within a pre-specified caliper. Greedy matching starts by creating two pools of patients; 1 pool for intervention patients and 1 pool for usual care. Each pool is randomly sorted, then for each intervention patient, we select the first usual care patient in the randomly sorted pool that has a propensity score within the pre-specified caliper. Once a match is made, it is never broken even if another closer match exists. Patients will be matched based on the propensity for intervention group using the propensity score estimated in the primary analysis. If there are no usual care patients with a propensity score within the caliper of a given intervention patient then that intervention patient is not included in the matched sample. We will match on the logit of the propensity score and use a caliper with a width of 0.2 times the standard deviation of the logit of the propensity score as suggested by Austin [7]. To estimate the intervention effect on non-persistence among the propensity matched sample we will fit a logistic regression model stratified by matched pair. Matching on the propensity score is expected to reduce most of the observed differences in patient case mix between the two groups so further adjustment is not necessary.
Random Intercepts for Hospital: We will fit a logistic regression model with adjustment for the same selected patient characteristics and same propensity score as in the primary analysis. The only modification is that we will account for within hospital clustering using random intercepts for hospitals instead of GEE.

6.2 MACE

Propensity matched analysis: We will use the same matched sample described in section 6.1 to estimate the intervention effect on MACE using a Cox proportional hazards model stratified by matched pair.
Random Intercepts for Hospital: We will fit a Cox proportional hazards model with adjustment for the same selected patient characteristics and same propensity score as in the primary analysis. The only modification is that we will account for within hospital clustering using random intercepts for hospitals instead of robust standard errors.

7. References

  1. Choudhry NK, Avorn J, Glynn RJ, Antman EM, Schneeweiss S, Toscano M, Reisman L, Fernandes J, Spettell C, Lee JL, Levin R, Brennan T, Shrank WH; Post-Myocardial Infarction Free Rx Event and Economic Evaluation (MI FREEE) Trial. N Engl J Med. 2011 Dec 1;365(22):2088-97
  2. Eldridge, S.M., D. Ashby, and S. Kerry, Sample size for cluster randomized trials: effect of coefficient of variation of cluster size and analysis method. Int J Epidemiol, 2006. 35(5): p. 1292-300.
  3. Federspiel JJ, Anstrom KJ, Xian Y, McCoy LA, Effron MB, Faries DE, Zettler M, Mauri L, Yeh RW, Peterson ED, Wang TY; Treatment With Adenosine Diphosphate Receptor Inhibitors-Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) Investigators. Comparing Inverse Probability of Treatment Weighting and Instrumental Variable Methods for the Evaluation of Adenosine Diphosphate Receptor Inhibitors After Percutaneous Coronary Intervention. JAMA Cardiol. 2016 Sep 1;1(6):655-65.
  4. Ho, P.M., et al., Multifaceted intervention to improve medication adherence and secondary prevention measures after acute coronary syndrome hospital discharge: a randomized clinical trial. JAMA Intern Med, 2014. 174(2): p. 186-93.
  5. Campbell M.K., Piaggio, G., Elbourne, D.R., Altman, D.G., Consort 2010 statement: extension to cluster randomized trials. BMJ, 2012. 345:e5661
  6. Austin, P.C., Balance diagnostics for comparing the distribution of baseline covariates between treatments groups in propensity-score matched samples. Statistics in Medicine, 2009. 28: p. 3083-3107.
  7. Austin, P.C., Some methods of propensity-score matching had superior performance to others: results of an empirical investigation and Monte Carlo simulations. Biometrical Journal, 2009. 51: p. 171-184.
Statistical Analysis Plan
8. Appendices
8.1 Appendix: Table of Contents for Statistical Tables, Figures, and Listings
Title Output
Site Characteristics by Copayment Intervention Table 1
Patient Level Baseline Characteristics by Copayment Intervention Table 2a
Cluster Level Baseline Characteristics by Copayment Intervention Table 2b
Clinical Outcomes by Copayment Intervention Table 3a
Subgroup Analysis of MACE by Copayment Intervention Table 3b
Medication Non-Persistence by Copayment Intervention Table 4a
Subgroup Analysis of Medication Non-Persistence by Copayment Intervention Table 4b
ICCs Table 5
Medication Use by Patient Reported Persistence and Copayment Intervention Table 6
Kappa Statistics for Assessing Agreement with Patient reported persistence to any at 1 year Table 7
Longitudinal Patterns of use Table 8
Kaplan-Meier Rates by Copayment Intervention Table 9a
Kaplan-Meier Rates of MACE by Copayment Intervention among subgroups Table 9b
Study Inclusion Criteria Figure 1
Study Inclusion Criteria under 1:1 randomization Figure 2a
Study Inclusion Criteria under 2:1 randomization Figure 2b
Kaplan-Meier Curves by Copayment Intervention Figure 3
Statistical Analysis Plan

8.2 Appendix: Table Shells

8.2.1 Main Table Shells

Table 1: Site Characteristics by Copayment Intervention
Overall ( ) Copayment Intervention ( ) Usual Care ( )
N patients enrolled
MI volume
% high ( ) MI volume
Site % baseline ticagrelor use
% high ( ) baseline ticagrelor use
Region
Randomization scheme
Total bed size
Teaching status
Government hospital
Member of a healthcare network
Surgery capabilities
Statistical Analysis Plan
Table 2a: Patient Level Baseline Characteristics by Copayment Intervention
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Age
Gender, % male
Non-white race
Hispanic ethnicity
Private insurance
Prior MI
Prior PCI
Prior CABG
Prior stroke/TIA
Prior heart failure
Dialysis
PAD
Hypertension
Diabetes
Current/recent smoker
Weight
Transfer in
STEMI
Home P2Y12 inhibitor
Home aspirin
Creatinine Clearance
Nadir hemoglobin
Multivessel disease
Access Site
PCI performed
CABG performed
Drug-eluting stent
In-hospital or prior bleeding
In-hospital MI
In-hospital stroke
Cardiogenic shock (Killip IV on presentation or inhospital cardiogenic shock)
Statistical Analysis Plan
Heart failure (Killip II/III on presentation or in-hospital heart failure)
Cardiac Arrest
Health Literacy
Baseline angina frequency
Cardiac Rehab
Baseline PHQ2>3
Baseline EQ5D VAS
Married
Employed
Education (college graduate)
Baseline financial hardship
Missed > 1 dose of medication in the last month
Statistical Analysis Plan
Table 2b: Cluster Level Summary Data of Baseline Characteristics by Copayment Intervention
Overall ( sites ) Copayment Intervention ( sites ) Usual Care ( sites )
Randomization Scheme
Site MI volume
Site % Ticagrelor
Age
Gender, % male
Non-white race
Hispanic ethnicity
Private insurance
Prior MI
Prior PCI
Prior CABG
Prior stroke/TIA
Prior Heart Failure
Dialysis
PAD
Hypertension
Diabetes
Current/recent smoker
Weight
Transfer in
STEMI
Home inhibitor
Home aspirin
Creatinine Clearance
Nadir hemoglobin
Multivessel disease
Access Site
PCI performed
CABG performed
Bare Metal Stent
In-hospital or prior bleeding
In-hospital MI
In-hospital stroke
Statistical Analysis Plan
Cardiogenic shock (Killip IV on presentation or inhospital cardiogenic shock)
Heart failure (Killip II/III on presentation or in-hospital heart failure)
Cardiac Arrest
Health Literacy
Baseline angina frequency
Cardiac Rehab
Baseline PHQ2>3
Baseline EQ5D VAS
Married
Employed
Education (college graduate)
Baseline financial hardship
Missed >1 dose of medication in the last month
Statistical Analysis Plan
Table 3a: Clinical Outcomes by Copayment Intervention
Cumulative Incidence at 12 months (95% CI) Unadjusted Adjusted
Outcome Overall ( ) Copayment Intervention ( ) Usual Care ( ) HR (95% CI) Pvalue HR (95% CI) Pvalue
MACE
All-cause death
MI
Stroke
BARC 2+ bleed
BARC 3+ bleed
GUSTO Moderate/Severe Bleed
Unplanned Revascularization
MACE + unplanned revascularization
Cardiovascular death
Statistical Analysis Plan
Table 3b: Subgroup Analysis of MACE by Copayment Intervention
Cumulative Incidence at 12 months (95% CI) Unadjusted Adjusted
Subgroup Overall ( ) Copayment Intervention ( ) Usual Care ( ) HR (95% CI) Pvalue HR (95% CI) Pvalue
Age
<65
Sex
Male
Female
Insurance type
Private
Non-private
Race
White
Non-white
MI type
STEMI
NSTEMI
In-hospital PCI
In-hospital PCI
No in-hospital PCI
Statistical Analysis Plan
Table 4a. Medication Non-Persistence by Copayment Intervention
Persistence at 12 months Unadjusted Adjusted
Outcome Overall ( ) Copayment Intervention ( ) Usual Care ( ) OR (95% CI) Pvalue OR (95% CI) Pvalue
Non-persistence to any vs. persistence
Non-persistence or mortality to any vs. persistence
Non-persistence to initial . persistence
Statistical Analysis Plan
Table 4b. Subgroup Analysis of Medication Non-Persistence to Any P2Y by Copayment Intervention
Persistence at 12 months Unadjusted Adjusted
Subgroup Overall ( ) Copayment Intervention ( ) Usual Care ( ) OR (95% CI) Pvalue OR (95% CI) Pvalue
Age
<65
Sex
Male
Female
Insurance type
Private
Non-private
Race
White
Non-white
MI type
STEMI
NSTEMI
In-hospital PCI
In-hospital PCI
No inhospital PCI
Statistical Analysis Plan
Table 5. ICCs
Overall Copayment Intervention Usual Care
Unadjusted ICC Adjusted ICC Unadjusted ICC Adjusted ICC Unadjusted ICC Adjusted ICC
MACE at 1 year
Nonpersistence to any vs. persistence
Medication Selection at discharge
Table 6. Medication Use by Patient Reported Persistence and Copayment Intervention in Substudies
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Persistence Measure Persistent Not Persistent Persistent Not Persistent Persistent Not Persistent
Fill Persistence by 1 year
Fill Adherence by 1 year
Drug levels at 1 year
Drug levels at 3 months
Drug levels at 6 months
Drug levels at 9 months
Table 7. Kappa Statistics for Assessing Agreement with Patient reported persistence to any at 1 year
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Persistence Measure Kappa (95% CI) Pvalue* Kappa (95% CI) P-value* Kappa (95% CI) Pvalue*
Fill Persistence by 1 year
Fill Adherence by 1 year
Drug levels at 1 year
Drug levels at 3 months
Drug levels at 6 months
Drug levels at 9 months
Statistical Analysis Plan

8.2.2 Supplementary Table Shells

Table 8: Longitudinal Patterns of use
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Overall Duration of any
Duration of initial
Summed duration of initial
Statistical Analysis Plan
Table 9a: Kaplan-Meier Cumulative Incidence
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Outcome Time (days)* KM rate (95% CI) N remaining N events KM rate (95% CI) N remaining N events KM rate (95% CI) N remaining N events
MACE
All-cause death
30
180
365
MI
Stroke
BARC 2+ bleed
BARC 3+ bleed
GUSTO
Moderate/Severe Bleed
30
180
365
Unplanned Revascularization
30
180
365
Cardiovascular death
Statistical Analysis Plan
180
365
MACE+Unplanned Revascularization
30
180
365
Statistical Analysis Plan
Table 9b: Kaplan-Meier Cumulative Incidence of MACE among Subgroups
Overall ( patients) Copayment Intervention ( patients) Usual Care ( patients)
Subgroup Time (days)* N events KM rate (95% CI) N remaining N events KM rate (95% CI) N remaining N events
Age>=65:
30
180
365
Age<65:
30
180
365
Male:
30
180
365
Female:
30
180
365
Private Insurance:
30
180
365
Government Insurance:
30
180
365
White:
30
180
365
Non-white:
30
Statistical Analysis Plan
365
STEMI:
30
180
365
NSTEMI:
30
180
365
In-hosp PCI:
30
180
365
No In-hosp
Statistical Analysis Plan

8.3 Appendix: List of Adjustment Variables

8.3.1 Adjustment Variables for Primary Models

Variable Variable Type
Intervention Yes/no
Randomization scheme Categorical (2:1 vs. 1:1 scheme)
Interaction* between intervention and randomization scheme Categorical
Site MI volume Categorical (high vs. low)
Site % Ticagrelor use Categorical (high vs. low)
Age Continuous
Male gender Yes/no
Race Categorical (white vs. nonwhite)
Insurance Payors Categorical (private vs. non-private)
Region Categorical (Northeast, West, South, vs. Midwest)
Propensity for intervention Continuous
Statistical Analysis Plan
8.3.2 Variables for Propensity Score Model
Variable Variable Type
Randomization scheme Categorical (2:1 vs. 1:1 scheme)
Age Continuous
Age >=65 vs. <65 Yes/no
Male gender Yes/no
Race Categorical (white vs. nonwhite)
Ethnicity Categorical (Hispanic vs. non-Hispanic)
Insurance Payors Categorical (private vs. non-private)
Prior MI Yes/no
Prior PCI Yes/no
Prior CABG Yes/no
Prior stroke/TIA Yes/no
Prior Heart failure Yes/no
Dialysis Yes/no
PAD Yes/no
Hypertension Yes/no
Diabetes Yes/no
Current/recent smoker Yes/no
Weight Continuous
Transfer in Yes/no
STEMI Yes/no
Home inhibitor Yes/no
Home aspirin Yes/no
Creatinine Clearance Continuous
Nadir hemoglobin Continuous
Multivessel disease Yes/no
Access Site Categorical (Femoral vs. other)
PCI performed Categorical (multivessel vs. culprit vs. none)
CABG performed Yes/no
Drug-eluting stent implanted Yes/no
In-hospital or prior bleeding Yes/no
In-hospital MI Yes/no
In-hospital stroke Yes/no
Cardiogenic shock (Killip IV on presentation or in-hospital cardiogenic shock) Yes/no
Heart failure (Killip II/III on presentation or inhospital heart failure) Yes/no
Cardiac Arrest Yes/no
Cardiac Rehab Referral Yes/no
Health Literacy Yes/no (score>=10 vs. <10)
Baseline angina frequency Categorical (100 vs. 70-90 vs. 0-60 points)
Baseline PHQ2>3 Categorical
Baseline EQ5D VAS Continuous
Statistical Analysis Plan
Married Yes/no
Employed Yes/no
Education (college graduate) Yes/no
Baseline financial hardship Categorical (1 vs. 2/3 vs. 4/5)
Missed > 1 dose of medication in the last month Yes/no
Site: Total bed size Continuous
Site: Teaching Status Yes/no
Site: Government hospital Yes/no
Site: Member of a Healthcare Network Yes/no
Site: Surgery Capabilities Yes/no

8.4 Appendix: Figure Shells

Figure 1: Consort Diagram
Statistical Analysis Plan
Figure 2a: Consort Diagram under 1:1 randomization
Statistical Analysis Plan
Figure 2b: Consort Diagram under 2:1 randomization
  1. *P-values are for test of Kappa (no more agreement than expected by chance).
  2. *Time (days) from discharge
  3. *Time (days) from discharge
  4. *We will test for significant interaction and will drop if the interaction term is not significant.