Assessment of Trends in Statin Therapy for Secondary Prevention of Atherosclerotic Cardiovascular Disease in US Adults From 2007 to 2016

Key Points Question What are the use, adherence rate, cost, and outcome trends in statin therapy for secondary prevention of atherosclerotic cardiovascular disease (ASCVD) in US adults from 2007 to 2016? Findings In this cohort study of data from 284 954 patients, modest increases in statin use, statin adherence rates, and cardiovascular outcomes among patients with ASCVD from 2007 to 2016 were found. The use of high-intensity statins approximately doubled over this time frame. Meaning Results of this study suggest that the increases in statin use, statin adherence rates, and cardiovascular outcomes were modest; and that substantial and persistent treatment gap exists.


eTable 1. Comparison of OptumLabs Data and a US Nationally Representative Cohort
To examine the temporal trends of statin use and outcomes without the influence of the change in baseline characteristics, the main analyses applied a weight so patient characteristics were similar across different years as demonstrated in eTable 3. eTable 6. Percentages of patients using high-intensity statins among patients who used statins within 30  Statin intolerance was assessed during the first year after the ASCVD event, defined using an established algorithm which includes: (1) down-titration of statin dose or switch to a lower-intensity statin, or (2) initiation of ezetimibe or a PCSK9 inhibitor within 7 days before or any time after or discontinuing statins, or (3) diagnosis for rhabdomyolysis or adverse effect of an antihyperlipidemic agent, or (4)  The purpose of this analysis plan is to provide guide to analysts when conducting the study. Most of the content will be included in the manuscript in order to guide researchers who want to replicate our findings or conduct similar studies. We also provided justifications for our methods and decisions so other researchers can make a choice or adjust their methods accordingly. Any time before the index date, used to establish a patient's medical history. Detailed description and justification can be found on page 12.
Patients can be follow up until 12/31/2017, which is the most recent data we will have at the time of the analysis. Please see page 19 for more details.

BACKGROUND AND OBJECTIVES
Atherosclerotic cardiovascular disease (ASCVD) is highly prevalent, affecting approximately 20 million people in the United States. 1, 2 Statin therapy is a key pillar of secondary prevention for these patients, 3 but numerous studies showed a significant rate of nonadherence and under-treatment. 4,5 As the benefits of adhering to statin therapy being repeatedly demonstrated, 6-8 the past decade has witnessed extensive efforts and innovative interventions to improve statin adherence. 9 During this time period, guidelines have been changed regarding high-intensity statins 10  As such, the current study will use a large national cohort of patients with all types of ASCVD managed at diverse routine practice settings to examine the use, adherence, cost, and outcomes of statin therapy for secondary prevention between 2007 and 2016. Specifically, the goal is to assess whether there have been improvements in treatment and outcomes and to identify gaps to guide future quality improvement efforts.

STUDY DESIGN AND DATA SOURCE
We will conduct a retrospective cohort analysis using OptumLabs Data Warehouse, which contains over 130 million privately insured and Medicare Advantage enrollees of all ages and races from all 50 states. 13

STUDY POPULATION
We will identify adult patients (age ≥21 years) who had their first ASCVD event between 1/1/2007-12/31/2016. The date of a patient's first ASCVD event will be defined as the index date. If a patient was hospitalized, the index date will be the discharge date from the hospital.
Patients will be required to have continuous medical and pharmacy insurance coverage for at least one year before the index date, to allow sufficient data to capture a relatively complete medical history. The time before the index date will be defined as the baseline. Based on our prior studies, patients on average have about 3-4 years of baseline period, which is stable over time, especially after adjusting for patient characteristics (e.g., age). This method of utilizing all baseline data available has been used in our previous studies. 15 Only patients who did not have ASCVD any time during the baseline period will be included.
ASCVD will be defined as MI, angina, coronary revascularization, ischemic stroke, TIA, or PAD. This definition is based on the 2013 American College of Cardiology/American Heart Association (ACC/AHA) guidelines. 10 We will identify MI, angina, ischemic stroke, and TIA, based on the presence of a primary diagnosis on an emergency room visit or a hospitalization. index date, in order to limit to ischemic stroke or TIA presumed to be of atherosclerotic origin. 16 Coronary revascularization will be identified using procedure codes. PAD will be identified using a validated algorithm with positive predictive value (PPV) of 91%. 17 In the validation study, PAD was defined as atherosclerotic occlusive arterial disease of the lower extremities, including arteries distal to the aortic bifurcation. The vascular laboratory diagnostic criteria for PAD were: (1) a resting/post-exercise ABI ≤0.9; or (2) the presence of poorly compressible arteries (ABI >1.4; or ankle blood pressure >255 mm Hg).
We will exclude patients who died or discontinued health insurance within the 30 days of discharge, patients whose hospitalization at the index date was more than 30 days, patients who received hospice care or stayed at a skilled nursing facility within 30 days of the index date, and patients who had invalid or missing demographic data.
The Mayo Clinic Institutional Review Board (IRB) exempted this study from review, because the study used pre-existing, de-identified data. We will use a broader set of codes to exclude those with a history of ASCVD at baselinewe will use 410.x, 411.0, 412, I22-I23, I24.1, and I25.2 for a history of MI and I69.3x for a history ischemic stroke, in addition to the codes in the above table. The diagnosis codes for CABG and PCI will be used for exclusion of patients with prior events only (i.e., they won't be used to identify a new event).

Baseline Characteristics
Baseline characteristics include socio-demographic characteristics, and medical history. Sociodemographic characteristics include age, sex, and race, determined at the time of index date.
Race/ethnicity is provided by OptumLabs, classified as non-Hispanic White (White), non-Hispanic Black (Black), Asian, Hispanic, or other/unknown. Self-report was the primary source, and when it was missing, imputation was made by the data provider based on other available administrative data. 18 Medical history will be determined using patients' physician, facility and pharmacy claims before the index date. All the comorbidities (except diabetes) will be defined as a diagnosis code or procedure code at any position in either inpatient or outpatient claims at baseline. Diabetes will be defined using an established algorithm, i.e., at least 2 outpatient encounters, or 1 inpatient encounter, or a antidiabetic prescription (not including metformin). 19 We will not use antihypertensive drugs to define hypertension, since some antihypertensive drugs can be used to treat other conditions.

Follow up
The end of the study period will be December 31, 2017. Patients will be followed until December 31, 2017, or the date of disenrollment in health insurance plans or death, whichever happened first.

Post-Discharge Statin Use
Post-discharge statin use will be defined as the use of statin within 30 days of discharge. The use of statin will be determined by the prescription fill in pharmacy claims. If a patient filled a statin prescription prior to the index date, and there were residual pills left (determined based on days supplied), the patient will be considered as on statin as well. As a sensitivity analysis, we will test the prescription fill within 90 days of discharge without considering prescription before the index date. We will calculate the proportion of patients who used prescription within 30 days of discharge. We will also calculate the proportion of statin users who received a high-intensity statin, defined as atorvastatin 40 or 80 mg, rosuvastatin 20 or 40 mg, and simvastatin 80 mg. 10 The FDA issued a warning against the use of simvastatin 80 mg in 2011, however, we will include this drug in the study since it was used early in the study period. As a sensitivity analysis, we will consider patients who received moderate-intensity statins plus ezetimibe in addition to high-intensity statins as high-intensity lipid-lowering treatment.
There is concern that whether the generic drug discount programs (e.g., Walmart's $4 drug plan) will affect the completeness of prescription information in the claims database.
However, in this study, the generic drug discount programs would have little impact on the completeness of the data. Frist, statins offered by drug discount programs are very limited. For example, Walmart and CVS only offer lovastatin 10 mg and 20 mg, which are low-intensity statins, but in patients with ASCVD, high-intensity statins (or moderate-intensity statin in patients with age>75 or safety concerns) are recommended. 10 Furthermore, lovastatin, regardless intensity, is not commonly used. In a previous study, only 10% of statin users received lovastatin. 12 Second, in patients with insurance, the out-of-pocket cost for a generic statin is very lowthe median was $2 for a 30-day supply in our dataset. Therefore, they may not need to turn to the $4 drug plan. Third, some previous studies assessing the completeness of claims data suggested that the payers have captured the claims when patients used a generic drug discount program. 20, 21

Statin Adherence
We will assess patients' adherence at one year, measured by the proportion of days covered (PDC). The analysis will be limited to patients who used statin within the first 30 days post discharge and continuously enrolled in health insurance during the 12 months after the index date. Since some patients were re-hospitalized during follow up, the number of days in a hospital will be deducted from the denominator when calculating PDC. Patients with a PDC≥80% will be considered as being adherent to statin therapy. When calculating PDC, all statins will be counted, even if the drug or the dose is different than what patients first received.

Statin Cost
Among patients who used statin within 30 days of discharge, we will calculate the cost of the statin therapy, standardized toward a 30-day supply (i.e., cost*30/days supplied). We will also adjust the cots to reflect the 2016 dollars using the Gross Domestic Product (GDP) price index, as recommended by the Agency for Healthcare Research and Quality (AHRQ). 22,23 We will calculate the amount paid by health plans, amount paid by patients, and the overall cost.

Cardiovascular Outcomes
We will assess the major adverse cardiac events (MACE), including MI, ischemic stroke, revascularization, and all-cause mortality during follow up. MI, ischemic stroke and revascularization will be defined the same way as how we identified the index events. As a sensitivity analysis, we will examine the lower extremity vascular complications, including surgical or endovascular procedures and amputations, in patients with PAD.

Statin Intolerance
The statin intolerance will be examined within one year after the index date. The analysis will be limited to patients who had continuous insurance coverage health insurance during the 12 months. The definition of statin intolerance is based on an established algorithm, 24, 25 including

Internal Validation of OLDW Data
The codes and algorithms used herein have been commonly used and validated in many previous studies. [26][27][28][29][30][31][32] We also leveraged the ability to link to laboratory results and electronic health records to validate our diagnosis codes. For example, we compared eGFR with the presence of a diagnosis code of Stage 3-4 chronic kidney disease (CKD) in those who did not have renal failure. We found 88% of patients who had a diagnosis of Stage 3-4 CKD had eGFR <60 mL/min/1.73m 2 , and 90% of those who did not have a diagnosis had eGFR ≥60 mL/min/1.73m 2 , which indicates good performance of the diagnosis codes. Moreover, the discrepancy between the diagnosis codes and eGFR could be because some patients may have a temporary decline in eGFR, but later recovered and did not develop to CKD or some patients had serum creatinine tests in facilities that did not submit data to OLDW.
We have also compared the ejection fraction documented in electronic health records and the diagnosis codes of heart failure. Using an ejection fraction cutoff of ≤40% for systolic heart failure diagnosis codes and ejection fraction of ≥50% for diastolic heart failure codes; we observed the specificity of 91% and 81%, respectively and sensitivity of 81% and 91%, respectively. We will include systolic heart failure in the propensity score model as a surrogate for reduced ejection fraction, but we acknowledge the inherent limitations in classification of heart failure by ejection fraction. 33 We have also conducted validation of the major bleeding diagnosis codes based on the International Society on Thrombosis and Haemostasis (ISTH) criteria 34 : (1) fatal bleeding, and/or, (2) symptomatic bleeding in a critical area or organ, such as intracranial, intraspinal, intraocular, retroperitoneal, intraarticular or pericardial, or intramuscular with compartment syndrome, and/or, (3) bleeding causing a fall in hemoglobin level of 2 g/dL or more, or leading to transfusion of two or more units of whole blood or red cells. We used ICD-9 and CPT procedure codes to identify transfusion, but we were not able to know the units of whole blood or red cells used in the transfusion. We also identified other procedures to control or manage bleeding, such as endoscopic procedures to address gastrointestinal bleeding, neurosurgical decompression for intracranial bleeding, evacuation of hematoma, or vascular embolization procedures to control bleeding. Among all bleeding events, one in four was bleeding in critical areas, and one third required transfusion. This is generally consistent with previous studies that adapted ISTH definition using administrative data. 35  however, this should be non-differential between treatment groups and should not influence our comparison. In fact, the mortality data is more reliable than most measures derived from administrative data, since its specificity is nearly perfect, and the sensitivity is also very high.

STATISTICAL METHODS
Based on our prior studies, patients in the early years (e.g., 2007-2010) were younger than those in later years, due to the increasing expansion of Medicare Advantage plans and increasing number of elderly patients in the OLDW database. We will use a weight to adjust for the baseline characteristics. The weight will be calculated from a multinomial logistic regression with the year of the index event as the outcome, and patient characteristics (in the above Table 3) as covariate. The weight will be one divided by the predicted probability of a patient's year of index date.
We will assess the trends of the outcomes in the overall ASCVD population as well as in the three subgroups defined based on a patient's index event: (1) MI, angina, and coronary revascularization; (2) ischemic stroke and TIA; (3) PAD. We will further assess the trends by age, sex and race/ethnicity. Logistic regression will be used to assess the trends of binary outcomes, including statin use, adherence and statin intolerance. Linear regression will be used to assess the trends of costs. Cox proportional hazards regression will be used to assess MACE, and the cumulative risk of MACE at the end of one year will be calculated using Kaplan -Meier survivor function.
A P value less than 0.05 will considered statistically significant for all tests. All tests will be 2-sided. All analyses will be conducted using SAS 9.4 (SAS Institute Inc.) and Stata 14.1 (Stata Corp).

LIMITATIONS
Our study relies on administrative data to ascertain baseline characteristics and outcomes, which could be subject to misclassification. However, it is unlikely there is any systematic difference in the ascertainment of comorbidities and outcomes between different treatment groups, and thus, the misclassification should not meaningfully impact our comparisons between drugs. The diagnosis and procedure codes used in this study have been commonly used in previous studies, and demonstrated good performance in our internal validation using linked laboratory results and electronic health records (described in Section 4.4) as well as other validation studies with positive predictive value around 90%. 17,26,[37][38][39] Second, our study will only include privately insured and Medicare Advantage patients.
The patient characteristics and outcomes could be different in the Medicaid, Medicare Fee-for-Service, and uninsured populations. However, the insurance coverage rates are high in older Americans. Over 90% of Americans aged 50-64 have health insurance and over 75% had private Although traditionally Medicare Advantage attracted healthier people, after the risk adjustment system was phased in from 2004-2007, the favorable risk selection has been largely reduced. 42 In fact, the results from this study will be more generalizable than most observational studies using other data sources. Observational studies largely use either administrative data or registries. Some cardiovascular registries focused on cardiology practices for recruitment and patients have to sign informed consent and agree to participate and to be actively followed, and thus the patients in these registries were more selective. Some administrative data are limited within a health system, within a region, or within an age range (e.g., Medicare, Kaiser, etc.). The OptumLabs Data Warehouse contains patients of all ages and races managed at heterogeneous practice settings from all 50 states. 13,14 The distribution of patient characteristics (e.g., age, sex and race/ethnicity) in the database is similar to those of the general U.S. population. 14 The data are updated monthly and are generally believed to be timely, accurate, and reflective of contemporary practice patterns. The concordance between OptumLabs and everyday practice is a major strength of the data source