Optimal P2Y₁₂ inhibitor durations in older men and older women following an acute myocardial infarction: A nationwide cohort study using Medicare data

Abstract

Study objective

Identify optimal P2Y₁₂ inhibitor durations balancing ischemic-benefit and bleeding-risk outcomes after acute myocardial infarction (AMI) in older men and women.

Design

Observational retrospective cohort with 2 years of follow-up, using clone-censor-weight marginal structural models to emulate randomization.

Setting

20 % sample of US Medicare administrative claims data.

Participants

P2Y₁₂ inhibitor new users ≥66 years old following 2008–2013 AMI hospitalization.

Exposures

12- to 24-month P2Y₁₂ inhibitor durations in 1-month intervals.

Main outcome measures

Effectiveness outcome (composite of all-cause mortality, recurrent AMI, ischemic stroke), safety outcome (hospitalized bleed), and negative control outcome (heart failure hospitalization).

Results

Of 28,488 P2Y₁₂ inhibitor new users, 51 % were female, 50 % were > 75 years old, 88 % were White/non-Hispanic, and 93 % initiated clopidogrel. Negative control outcome results for 16- through 24-month durations appeared most likely to meet assumptions of no unmeasured confounding. Compared to men taking 24-month therapy, men taking 16-month therapy had higher 2-year risks of the composite effectiveness outcome (relative risk [RR] = 1.08; 95 % confidence interval [95%CI]:1.00–1.15) with similar bleeding risks (RR = 0.98; 95%CI:0.85–1.13). Compared to women taking 24-month therapy, women taking 16-month therapy had similar 2-year risks of the composite effectiveness outcome (RR = 0.98; 95%CI:0.92–1.04) and lower bleeding risks (RR = 0.88; 95%CI:0.80–0.96).

Conclusions

Older men taking 24-month P2Y₁₂ inhibitor therapy had the lowest composite effectiveness outcome risk with no increased bleeding risk compared to shorter durations. Women taking 16-month versus 24-month P2Y₁₂ inhibitor therapy had similar composite effectiveness outcome risks but a substantially lower hospitalized bleeding risk, suggesting durations beyond 15–17 months lacked benefit while increasing bleeding risk.

Highlights

• •Nationwide observational cohort study in Medicare beneficiaries ≥66 years old
• •Effectiveness and safety of P2Y₁₂ inhibitor durations evaluated in older patients
• •Novel clone-censor-weight approach emulated randomization to therapy durations.
• •In older men, durations up to 24 months beneficial without increasing bleeding risk
• •In older women, durations >15 = 17 months increased bleeding risk without benefit.

1. Introduction

Dual antiplatelet therapy (DAPT)—low-dose aspirin plus a P2Y₁₂ inhibitor—has demonstrated benefit in preventing major adverse cardiovascular and cerebrovascular events in randomized controlled trials (RCTs) and is often part of guideline-recommended secondary prevention for acute myocardial infarction (AMI) [1]. When post-AMI DAPT is indicated, 2016 American College of Cardiology/American Heart Association guidelines recommend continuing aspirin indefinitely, but in most scenarios, P2Y₁₂ inhibitor therapy may be discontinued after “at least” 12 months to reduce the risk of bleeding [1]. The ischemic-benefit/bleeding-risk tradeoff has led to considerable debate regarding the optimal therapy duration, with 12 RCTs published since 2010 directly comparing anywhere from 3- up to 48-month P2Y₁₂ inhibitor durations [[2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13]].

Differences between women and men exist in AMI pathophysiology and symptom presentation [14,15]. Therefore, women and men may respond differently to secondary prevention medications. When comparing longer to shorter P2Y₁₂ inhibitor durations for composite major adverse cardiovascular and cerebrovascular events, 7 RCTs reported subgroup results by sex: 1 found statistically significant results favoring longer vs. shorter durations in men [12], 5 others also had non-significant estimates favoring longer durations in men [[6], [7], [8], [9], [10]], and 6 RCTs reported either non-significant estimates favoring shorter durations or found no difference in women [[6], [7], [8], [9], [10], [11]].

While statistical tests to detect subgroups interactions within each of these trials may not individually identify a difference in results between men and women (comparisons for which they were not powered to detect), the overall pattern of findings across these 7 RCTs suggest men may be more likely than women to benefit from longer P2Y₁₂ inhibitor durations for the prevention of composite major adverse cardiovascular and cerebrovascular events. Some composite outcomes in these RCTs included bleeding and one study found that men and women experienced a similar increase in the rate of bleeding when comparing 30- vs. 12-month P2Y₁₂ inhibitor therapy [12], but most RCTs did not report bleeding safety outcomes separately for men and women. A meta-analysis of 6 RCTs comparing longer DAPT durations (12 months) vs. shorter DAPT durations (3 or 6 months) found that the comparative 1-year rate of bleeding was similar in men and women (P-interaction = 0.25), but men could potentially be more likely than women to benefit from longer durations for the comparative 1-year rate of major adverse cardiac events (P-interaction = 0.08) [16].

A major limitation of the existing evidence comparing the effectiveness of P2Y₁₂ inhibitor durations is that much of the US Medicare population is underrepresented, with an average age < 65 in 9 of 12 RCTs [[2], [3], [4], [5],[8], [9], [10], [11], [12]]. Additionally, of 36,221 participants in these 12 RCTs, only 27 % were women. Of note, women often have their first AMI later in life and account for ~50 % of Medicare AMI hospitalizations [[17], [18], [19], [20]]. The optimal duration of P2Y₁₂ inhibitor therapy is largely unknown among older adults in real-world settings, especially among older women given their bleeding risk may be higher [1,21,22]. Therefore, we aimed to identify the optimal P2Y₁₂ inhibitor duration following an AMI hospitalization among men and women 66 years of age and older using a novel clone-censor-weight marginal structural modeling approach, emulating a trial that randomized patients to different therapy durations.

2. Materials and methods

The University of North Carolina at Chapel Hill institutional review board approved and waived informed consent requirements for this secondary analysis of administrative claims data. A 20 % sample of 2007–2015 fee-for-service Medicare claims (enrollment summary, medical service claims, and Part D prescription claims) was the primary data source [23]. Our data use agreement with the Centers for Medicare and Medicaid Services (CMS) prohibits sharing research data; access to Medicare administrative claims data for research can be requested through application with CMS. Prescriber and contextual characteristics were identified by linking to previously described data sources [19].

Patients with 2008–2013 AMI hospitalizations (primary/secondary position International Classification of Diseases, Ninth Revision [ICD-9] code 410.x1) [24] were eligible if they (1) were ≥ 66 years old; (2) had no recent P2Y₁₂ inhibitor use/indication in 12 months pre-AMI; (3) filled a P2Y₁₂ inhibitor prescription between hospital admission and 30 days post-discharge; and (4) did not discontinue P2Y₁₂ inhibitor therapy before a second prescription fill (Fig. 1). Restricting our sample to only patients who filled a P2Y₁₂ inhibitor (i.e., had an indication without major contraindications) helps mitigate unmeasured confounding in observational research [25]. This study's sample is a subset of our previously described P2Y₁₂ inhibitor new-user cohort, where the only missing baseline variable was age of the index P2Y₁₂ inhibitor prescriber (1.3 % missing and stochastically imputed) [19,26,27].

Fig. 1.

P2Y₁₂ inhibitor new user cohort study flowchart. Abbreviations: AMI, acute myocardial infarction; CABG, coronary artery bypass surgery; ESRD, end-stage renal disease; HIV, human immunodeficiency virus; PCI, percutaneous coronary intervention; SNF, skilled nursing facility.

A retrospective cohort design was used (Fig. 2). Follow-up occurred in 30-day intervals (i.e., person-months) beginning in month 2 (days 31–60 post-AMI), and continued until experiencing an outcome, censoring, or end of follow-up after 24 months (2 years).

Fig. 2.

Study design. Index AMI hospitalization (A) from date of admission (0_i,adm) to discharge (0_i,dis). Baseline conditions and previous P2Y₁₂ inhibitor use/indication (B) and use of other medications (C) measured pre-AMI. P2Y₁₂ inhibitor prescription fill and event-free survival required during first 30 days post-AMI (D). Follow-up (E) began in month 2 (day 31) and occurred in 30-day increments (person-months).

Abbreviations: AMI, acute myocardial infarction; CABG, coronary artery bypass surgery; PCI percutaneous coronary intervention.

2.1. Exposure

P2Y₁₂ inhibitor therapy duration in months was the exposure. We compared results for 13 different regimens (in 1-month intervals) of 12 to 24 continuous months of P2Y₁₂ inhibitor therapy without discontinuation. Medication stop dates are typically unavailable in administrative prescription claims data. To differentiate medication discontinuation from continuous use with suboptimal adherence (e.g., nonadherence from missed doses every other day), a “gap days” approach was used [19,28]. Medication availability is assessed during each day in the measurement period, and once a period of consecutive days (e.g., 30 days) have passed where the patient had no medication available and did not fill a prescription to replenish their supply, they are classified as discontinuing their medication. If a patient filled a 30-day supply prescription and a “gap days” measure of 30 was used to determine the discontinuation date, that patient would have a proportion of days covered (PDC) medication adherence of 50 % on the discontinuation date if they were actually continuing to use the medication with suboptimal adherence but were misclassified as discontinuing the medication [30d prescription/(30d prescription + 30d gap day) = 50 %].

To operationalize this in our study, P2Y₁₂ inhibitor availability was assessed each day of follow-up using prescription fill dates and days supply from Part D prescription claims, accounting for oversupply (e.g., filling prescriptions a few days early so a patient's supply of medication does not run out) and hospitalizations (medications normally taken at home are provided by the hospital, so a patient's home supply would not be used during this period) [19,20,29]. The last day of a 30-day drug-free interval was the stop date (i.e., 30 consecutive days without therapy) [19]. The month of discontinuation was determined from this stop date (e.g., month 14 if discontinued between 391 and 420 days). The robustness of estimates when using our 30-day “gap days” measure to determine the discontinuation date was evaluated in sensitivity analyses (see Section 2.4.3 below).

2.2. Outcomes

The effectiveness outcome was a composite of (1) all-cause mortality, (2) recurrent AMI [24], and (3) ischemic stroke (Table 1) [[30], [31], [32], [33], [34], [35], [36], [37]]. In previously studied Medicare populations, this AMI algorithm had a positive predictive value (PPV) of 94.1 % [24], and this ischemic stroke algorithm had a PPV of 88.6 % and specificity of 99.8 % [36].

Table 1.

ICD-9 codes and algorithms used to measure study outcomes.

Outcome		ICD-9 codes	Source of diagnosis code	Diagnosis field position
Composite clinical effectiveness outcome (all-cause mortality and hospitalized ischemic events)
All-cause mortalitya		Date of death	Beneficiary summary file
Recurrent AMI [24]		410.x1	Inpatient	Primary or secondary
Ischemic stroke [[30], [31], [32], [33], [34], [35], [36], [37]]		433.x1, 434 (excluding 434.x0), 436	Inpatient	Primary
Clinical safety outcome (hospitalized bleeding events)
Intracranial hemorrhage [[30], [31], [32], [33], [34], [35],[38], [39], [40], [41]]		430, 431, 432	Inpatient	Primary
Gastrointestinal hemorrhage [32,33,[38], [39], [40],42]		455.2, 455.5, 455.8, 456.0, 456.20, 530.7, 530.82, 531.0–531.6, 532.0–532.6, 533.0–533.6, 534.0–534.6, 535.01, 535.11, 535.21, 535.31, 535.41, 535.51, 535.61, 537.83, 562.02, 562.03, 562.12, 562.13, 568.81, 569.3, 569.85, 578.0, 578.1, 578.9	Inpatient	Primary
Other bleeding events [32,33,[38], [39], [40]]			Inpatient	Primary
	Choroidal hemorrhage	363.6
	Conjunctival hemorrhage	372.72
	Orbital hemorrhage	376.32
	Optic nerve hemorrhage	377.42
	Vitreous hemorrhage	379.23
	Hemopericardium	423.0
	Hematuria	599.7
	Hemarthrosis	719.1
	Hematoma	729.92
	Adrenal hemorrhage	772.5
	Epistaxis	784.7
	Hemorrhage from throat	784.8
	Hemoptysis	786.3
	Injury to kidney	866.01–866.02, 866.11–866.12
	Other	285.1, 459.0
Negative control outcome (hospitalized heart failure)
Heart failure [[45], [46], [47], [48]]		428.x, 402.01, 402.11, 402.91	Inpatient	Any

Abbreviations: AMI, acute myocardial infarction; ICD-9, International Classification of Diseases, Ninth Revision, Clinical Modification.

^aMeasured using date of death in annual beneficiary summary file.

The safety outcome was a hospitalized bleed, including intracranial hemorrhages [[30], [31], [32], [33], [34], [35],[38], [39], [40], [41]], gastrointestinal hemorrhages [32,33,[38], [39], [40],42], and other bleeding events (Table 1) [32,33,[38], [39], [40]]. Our hospitalized bleed outcome measure was primarily developed from an algorithm with PPV 89 % to identify clinically confirmed definite or probable bleeding at any anatomical site as the primary reason for hospitalization among anticoagulant users [38]. Similar algorithms had a PPV of 92 % with diagnosis codes in the primary position only to identify hospitalization for bleeding of any severity level in patients treated with P2Y₁₂ inhibitors after receiving PCI for an AMI [43].

Additionally, a negative control outcome was used to assess the potential impact of unmeasured confounding (Table 1) [44]. Heart failure hospitalizations [[45], [46], [47], [48]] were chosen as the negative control outcome because (1) in the absence of residual confounding, no association was expected between P2Y₁₂ inhibitor duration and risk of heart failure hospitalizations, and (2) potential unmeasured confounders in our data source (e.g. healthy-user/sick-stopper characteristics and cardiovascular disease severity) between P2Y₁₂ inhibitor duration and clinical effectiveness/safety outcomes are also potential unmeasured confounders between P2Y₁₂ inhibitor duration and heart failure hospitalizations [44,49,50]. The ICD-9 codes for our negative control outcome had a PPV of 97 % to identify incident heart failure hospitalizations when used in the primary diagnosis field [45], but we expanded our search to these ICD-9 codes in any position to maximize our sensitivity to identify acute decompensated or chronic stable heart failure hospitalizations [47], as neither were expected to be associated with duration of P2Y₁₂ inhibitor therapy in the absence of unmeasured confounding.

2.3. Covariates

Our research framework and baseline covariates (V in equations)—including pre-AMI and index AMI characteristics as well as contextual characteristics—were previously described [19,51]; however, in this study, variables measured 30 days post-AMI were included in time-varying covariates, not baseline covariates. Patient baseline characteristics included sociodemographics, healthcare/medication utilization measures, comorbidity burden, specific comorbidities/medications that could impact ischemic/bleeding risks, and medications with P2Y₁₂ inhibitor drug-drug interactions.

The following time-varying covariates (L in equations) were measured: P2Y₁₂ inhibitor out-of-pocket cost, deductible/coverage-gap Part D benefit phase, proton-pump inhibitor use, and hospice admission. Time-varying covariates attributed to time t were measured in time period t-1 to maintain correct confounder-exposure-outcome temporal ordering [[52], [53], [54], [55]] in our clone-censor-weight analysis described below (e.g., when estimating probability of remaining uncensored during month 14 of follow-up, time-varying covariates were measured at the end of month 13).

Time-dependent effects (E in equations) attributed to time t were similarly measured at time t-1. Time-dependent effects were interacted with time when calculating inverse probability weights in our clone-censor-weight analysis described below; the probability of remaining uncensored and event-free was hypothesized to depend upon when these events occurred within the timeframe of post-AMI care (e.g., first few months vs. >1 year after index AMI hospitalization) [20,[56], [57], [58]]. Time-dependent effects were inpatient and emergency room admissions; outpatient primary care/cardiologist visits; use of other AMI secondary prevention medications; P2Y₁₂ inhibitor product (brand-name clopidogrel, generic clopidogrel, brand-name prasugrel, brand-name ticagrelor); coronary revascularization; and P2Y₁₂ inhibitor prescriber characteristics (sex, age, specialty). For each non-P2Y₁₂ inhibitor AMI secondary prevention medication (statin, angiotensin-converting enzyme [ACE] inhibitor/angiotensin II receptor blocker [ARB], beta-blocker), use during each period was classified as either not taking, taking but nonadherent (PDC <0.80), or taking and adherent (PDC ≥0.80). Additionally, AMI/ischemic stroke and bleeding hospitalizations were time-dependent effects when they were not the outcome.

See Supplemental Tables S1-S2 for a full list of all covariates. Additional details can be found in the footnotes of these tables regarding how covariates were measured; Part D benefit phase, P2Y₁₂ inhibitor prescription fill characteristics, and P2Y₁₂ inhibitor prescriber characteristics used the last value carried forward if no information was available within the 30-day t-1 time period of measure.

2.4. Statistical analysis

Covariate distributions were described separately for men and women. To overcome immortal-time bias and selection bias when estimating clinical effectiveness and safety of treatment durations in real-world observational data, we used a 3-step marginal structural modeling approach: cloning, censoring, and weighting [59,60]. This approach was applied separately for all 3 outcomes and estimates are interpreted similarly to an RCT per-protocol analysis that adjusts for time-varying confounding. When clone-censor-weight marginal structural models are successfully implemented to emulate an RCT, (1) there should be little to no difference in the risk of hospitalized heart failure (negative control outcome) among assigned P2Y₁₂ inhibitor durations, and (2) longer P2Y₁₂ inhibitor durations should have similar or higher risk of hospitalized bleed (positive control outcome) than shorter durations [44,61]. We decided a priori that when cumulative incidence curves aligned with these expected findings for negative and positive control outcomes, P2Y₁₂ inhibitor durations would be identified as comparable in our study, being more likely to meet the causal inference exchangeability assumption [62]. Estimates for P2Y₁₂ inhibitor durations that did not align with these expected findings for negative and positive control outcomes were interpreted cautiously because there may be unmeasured confounding or even a lack of clinical equipoise for patients receiving these P2Y₁₂ inhibitor durations in real-world clinical practice, thereby violating the exchangeability assumption [44,63].

All analyses were conducted using SAS 9.4 (SAS Institute Inc).

2.4.1. Marginal structural modeling using cloning, censoring, and weighting

To emulate randomization, each patient's data was copied 13 times, and each of these 13 “clones” was assigned to remain on P2Y₁₂ inhibitor therapy continuously from 12 to 24 months in 1-month intervals (i.e., 12-, 13-, 14-, …, 24-month durations). Assignment to these 13 therapy durations was analogous to randomization at AMI discharge [59,64].

Once a patient's actual P2Y₁₂ inhibitor duration deviated from their clone's “randomly” assigned duration, the clone was censored [59]. Clones were given a 1-month grace period to discontinue therapy (e.g., clones assigned 14-month therapy were not censored if patient actually discontinued in months 13–15, but they must remain off therapy for all months after discontinuation had occurred) [65,66]. Deviations from assigned therapy durations were measured using 3 censoring mechanisms [52]: stopping therapy early (C^stop in equations), stopping therapy appropriately but later restarting (C^restart in equations), and continuing therapy too long (C^cont in equations). Clones were also censored from loss to follow-up (C^abd in equations) if they (1) lost Medicare fee-for-service continuous enrollment, or (2) had 2 consecutive months with no P2Y₁₂ inhibitor available but had not yet discontinued (immeasurable inpatient time for adherence-related prescription claim measures) [29]. Finally, clones were censored when death was a competing risk in safety and negative control outcome analyses (C^death in equations).

Inverse probability weights (IPWs) were calculated for the probability of remaining uncensored through month t to eliminate time-varying selection bias introduced from informative censoring, constructing a hypothetical pseudo-population where all patients follow their assigned treatment regimen with no censoring [59,67]. Unstabilized IPWs were calculated using person-month pooled logistic regression [54,55,65] (Supplemental Eq. S1) and truncated at 1st and 99th percentiles [66,68].

2.4.2. Cumulative incidence curves

After calculating IPWs, pooled logistic regression was used to calculate event-free survival probabilities through month t; cumulative incidence curves were calculated—separately for men and women—using weighted survival probabilities and the product-limit method (Supplemental Eq. S2) [52,65,69]. Then, 2-year risks, relative risks (RRs), and risk differences (RDs) were calculated, comparing shorter P2Y₁₂ inhibitor durations to the 24-month reference duration. The goal of our analyses was not to determine statistical significance for the different P2Y₁₂ inhibitor durations and outcomes under study; instead, we used confidence intervals (CIs) to represent the precision and uncertainty of our estimated effect sizes for scientific inference [70,71]. The 95 % CIs were calculated with 500-iteration nonparametric bootstrapping with replacement from the original cohort, followed by cloning, censoring, and weighting within each iteration [53,65,69].

2.4.3. Sensitivity analyses

Two sets of sensitivity analyses were conducted. First, the 30-day drug-free interval to measure P2Y₁₂ inhibitor discontinuation was varied to shorter (15-day) and longer (45-day) intervals [19,28]. Second, analyses were stratified by calendar date of the index AMI discharge based on availability of P2Y₁₂ inhibitor products in the US market [19]. Point estimates from sensitivity analyses that fell within primary analyses' 95 % CIs were identified as consistent with primary findings.

3. Results

Overall, 28,488 P2Y₁₂ inhibitor new users (51 % women) were eligible for effectiveness and safety analyses. Among all 341,079 P2Y₁₂ inhibitor prescription fills from index AMI admission through 2 years post-AMI discharge in these 28,488 patients, 82 % and 15 % of prescription fills were for 30-day and 90-day supply, respectively. In negative control outcome analyses, 1581 patients with heart failure hospitalizations within 30 days post-AMI were excluded. See Online Supplemental Materials for these results: person-month distribution of all covariates in original, cloned, and weighted samples (Supplemental Tables S1-S2); crude outcome event rates (Supplemental Table S3); details on analytic samples after cloning and censoring (Supplemental Tables S4-S6); distribution of IPWs before truncating (Supplemental Table S7); and unadjusted and baseline-covariate-adjusted clone-censor-weight marginal structural model results (Supplemental Figs. S1-S2 and Supplemental Tables S8-S13).

3.1. Results for men

Of 13,920 men in effectiveness and safety analyses, 41 % were ≥76 years old, 90 % were White/non-Hispanic, 16 % had history of bleeding, 75 % had index AMI managed with PCI and stent(s), and 92 % initiated clopidogrel (Table 2; see Supplemental Table S14 for all baseline characteristics).

Table 2.

Selected baseline characteristics in male and female patients.

Patient baseline characteristics	Male patients	Female patients
	N = 13,920	N = 14,568
	No. (%)a	No. (%)a
Predisposing patient characteristics
Age
66–75	8222 (59.1)	6125 (42.0)
76–85	4467 (32.1)	5870 (40.3)
86+	1231 (8.8)	2573 (17.7)
Race/ethnicity
White, non-Hispanic	12,498 (89.8)	12,670 (87.0)
Black, non-Hispanic	626 (4.5)	1059 (7.3)
Hispanic	298 (2.1)	393 (2.7)
Asian, non-Hispanic	314 (2.3)	297 (2.0)
Other	184 (1.3)	149 (1.0)
Enabling patient characteristics
Dual Medicare/Medicaid eligibilityb	1993 (14.3)	3224 (22.1)
Patient-perceived need
Pre-AMI conditionsc
Charlson comorbidity index
0	5460 (39.2)	5191 (35.6)
1–2	5374 (38.6)	6158 (42.3)
3–4	2105 (15.1)	2236 (15.3)
5–7	753 (5.4)	766 (5.3)
8+	228 (1.6)	217 (1.5)
Depression	1075 (7.7)	2120 (14.6)
Dementia/Alzheimer'sd	605 (4.3)	1015 (7.0)
Pre-AMI medicationse
Chronic medicationsf
0	2012 (14.5)	1194 (8.2)
1–3	5087 (36.5)	4328 (29.7)
4–6	4062 (29.2)	4797 (32.9)
7–9	1868 (13.4)	2684 (18.4)
10+	891 (6.4)	1565 (10.7)
Prescriber-perceived need
Pre-AMI conditionsc
Diabetes	4404 (31.6)	4870 (33.4)
Heart failure	1504 (10.8)	1949 (13.4)
Coronary artery disease	4791 (34.4)	4068 (27.9)
Cerebrovascular disease	509 (3.7)	560 (3.8)
Peripheral vascular disease	1542 (11.1)	1893 (13.0)
Cancer	2088 (15.0)	1178 (8.1)
Previous venous thromboembolism	865 (6.2)	941 (6.5)
History of smokingg	5443 (39.1)	3983 (27.3)
Atrial fibrillation	974 (7.0)	1043 (7.2)
History of bleeding event	2201 (15.8)	2061 (14.1)
Pre-AMI medicationse
Statin	6081 (43.7)	6091 (41.8)
ACE inhibitor or ARB	6345 (45.6)	7645 (52.5)
Beta-blocker	5461 (39.2)	6458 (44.3)
Anticoagulant	1017 (7.3)	974 (6.7)
Index AMI hospitalization characteristics
Year of index admission
2008	2334 (16.8)	2657 (18.2)
2009	2254 (16.2)	2493 (17.1)
2010	2304 (16.6)	2547 (17.5)
2011	2453 (17.6)	2519 (17.3)
2012	2539 (18.2)	2510 (17.2)
2013	2036 (14.6)	1842 (12.6)
AMI intervention strategyh
PCI with drug-eluting stent	7307 (52.5)	7004 (48.1)
PCI with bare-metal stent	3099 (22.3)	2888 (19.8)
Other PCI	893 (6.4)	716 (4.9)
Coronary artery bypass surgery	543 (3.9)	282 (1.9)
Medical management or fibrinolytics	2078 (14.9)	3678 (25.2)
Heart failure	3278 (23.5)	4187 (28.7)
Bleeding event	918 (6.6)	851 (5.8)
Acute kidney injury	1477 (10.6)	1486 (10.2)
Index P2Y12 inhibitor producti
Brand-name clopidogrel	9822 (70.6)	10,794 (74.1)
Generic clopidogrel	2928 (21.0)	3066 (21.0)
Brand-name prasugrel	905 (6.5)	490 (3.4)
Brand-name ticagrelor	265 (1.9)	218 (1.5)

Abbreviations: ACE, angiotensin-converting enzyme; AMI, acute myocardial infarction; ARB, angiotensin II receptor blocker; ATC, Anatomical Therapeutic Chemical Classification System; PCI, percutaneous coronary intervention.

^aColumn percentages of total male or female patients included in cohort.

^bMedicare beneficiaries dually enrolled with full Medicaid benefits at any point in 12 calendar months before through 1 calendar month after index AMI hospitalization.

^cMeasured 12 months pre-AMI admission date.

^dMedicare Chronic Conditions Data Warehouse definition.

^eMeasured 6 months pre-AMI admission date.

^f≥2 prescription fills on separate dates for unique 4th-level ATC code.

^gMeasured 12 months pre-AMI admission date through 30 days post-AMI discharge date using Desai algorithm including diagnoses codes and prescription medication fill data.

^hMeasured during index AMI hospitalization; includes coronary revascularization procedures through 30 days post-AMI discharge date. Variable is hierarchical with categories officially being (1) any PCI with drug-eluting stent; (2) PCI with bare-metal stent but no drug-eluting stent; (3) PCI with no mention of coronary stent; (4) patients undergoing coronary artery bypass surgery but did not receive PCI; and (5) patients who (a) did not receive PCI, (b) did not receive coronary artery bypass surgery, and (c) either received fibrinolytics or had no record of other intervention strategies.

ⁱMeasured between index AMI admission date and 30 days post-AMI discharge date. Variable not included in clone-censor-weight marginal structural models; presented here to simply describe the percentage of patients filling each type of product as their first P2Y₁₂ inhibitor.

Men taking 24-month therapy had the lowest 2-year risk of death, recurrent AMI, or ischemic stroke (18.5 %; 95 % CI: 17.5, 19.5; Fig. 3A). Men taking the shortest duration of 12 months had the lowest 2-year bleed risk (4.4 %; 95 % CI: 3.8, 5.2; Fig. 3B). However, hospitalized bleed cumulative incidence curves sometimes did not separate until 9+ months after therapy discontinuation, and estimates were imprecise with wide confidence intervals. Men taking between 16- and 24-month P2Y₁₂ inhibitor durations had similar 2-year heart failure hospitalization risks (19.0–19.7 %; Fig. 3C); these durations were identified as most comparable for causal inference in men.

Fig. 3.

Effectiveness, safety, and negative control outcome cumulative incidences of P2Y₁₂ inhibitor durations in men. Estimates are presented for all durations, but cumulative incidence curves are only presented in 3-month intervals (12-, 15-, 18-, 21-, and 24-month durations). Lighter colors are shorter P2Y₁₂ inhibitor durations; darker colors are longer durations. Estimates (95 % CIs) for composite effectiveness (A), bleeds (B), and heart failure (C) outcomes. For (A) and (B), duration with lowest risk is bolded; for (C), comparable durations are bolded. Please note, the y-axis risk scale for bleeds (B) is different than other outcomes.

Abbreviations: AMI, acute myocardial infarction; CI, confidence interval; RD risk difference; RR, relative risk.

3.2. Results for women

Of 14,568 women in effectiveness and safety analyses, 58 % were ≥76 years old, 87 % were White/non-Hispanic, 14 % had history of bleeding, 68 % had index AMI managed with PCI and stent(s), and 95 % initiated clopidogrel (Table 2; see Supplemental Table S14 for all baseline characteristics).

Women taking 14-month P2Y₁₂ inhibitor therapy had the lowest 2-year risk of death, recurrent AMI, or ischemic stroke (20.1 %; 95 % CI: 18.9, 21.1; Fig. 4A). Women taking the shortest duration of 12 months had the lowest 2-year bleed risk (5.4 %; 95 % CI: 4.6, 6.4; Fig. 4B). Compared to women who continued therapy for longer durations and had higher bleed risks, the risk of bleeding began to decrease within the month following P2Y₁₂ inhibitor discontinuation. Women taking between 16- and 24-month P2Y₁₂ inhibitor durations had similar 2-year heart failure hospitalization risks (23.7–25.2 %; Fig. 4C); these durations were identified as most comparable for causal inference in women.

Fig. 4.

Effectiveness, safety, and negative control outcome cumulative incidences of P2Y₁₂ inhibitor durations in women. Estimates are presented for all durations, but cumulative incidence curves are only presented in 3-month intervals (12-, 15-, 18-, 21-, and 24-month durations). Lighter colors are shorter P2Y₁₂ inhibitor durations; darker colors are longer durations. Estimates (95 % CIs) for composite effectiveness (A), bleeds (B), and heart failure (C) outcomes. For (A) and (B), duration with lowest risk is bolded; for (C) comparable durations are bolded. Please note, the y-axis risk scale for bleeds (B) is different than other outcomes.

Abbreviations: AMI, acute myocardial infarction; CI, confidence interval; RD risk difference; RR, relative risk.

3.3. Summary of effectiveness and safety results in men and women

Magnified cumulative incidence curves for 16- to 24-month P2Y₁₂ inhibitor durations show shorter vs. longer durations were less effective at preventing the composite effectiveness outcome in men (Fig. 5A), but no difference was observed in women (Fig. 5B). The 2-year composite effectiveness estimates comparing 16-month vs. 24-month durations in men were RR = 1.08 (95 % CI: 1.00, 1.15) and RD = +1.4 % (95 % CI: +0.0, +2.6); in women the 16-month vs. 24-month RR = 0.98 (95 % CI: 0.92, 1.04) and RD = -0.4 % (95 % CI: −1.7, +0.8).

Fig. 5.

Magnified cumulative incidences of 16- to 24-month P2Y₁₂ inhibitor durations for the composite effectiveness outcome in men (A) and women (B), and the safety bleeding outcome in men (C) and women (D). Results for only 16- to 24-month P2Y₁₂ inhibitor durations (identified as comparable for causal inference) are presented here. Please note, the x-axis time scale goes from 15 to 24 months after AMI, and the y-axis risk scales are different for effectiveness vs. safety outcomes. Abbreviations: AMI, acute myocardial infarction.

There were no differences in the risk of the bleeding safety outcome in men attributable to P2Y₁₂ inhibitor therapy durations (Fig. 5C), but shorter durations were substantially safer than longer durations in women (Fig. 5D). The 2-year hospitalized bleeding estimates comparing 16-month vs. 24-month durations in men were RR = 0.98 (95 % CI: 0.85, 1.13) and RD = -0.1 % (95 % CI: −0.8, +0.6); in women the 16-month vs. 24-month RR = 0.88 (95 % CI: 0.80, 0.96) and RD = -0.7 % (95 % CI: −1.3, −0.2).

3.4. Sensitivity analyses

When varying the P2Y₁₂ inhibitor discontinuation measure to 15- and 45-day drug-free intervals, all estimates among patients taking between 16- and 24-month P2Y₁₂ inhibitor durations were consistent with estimates from primary analyses. When analyses were stratified by index AMI discharge date, several estimates were inconsistent with primary analyses; however, these inconsistent estimates were mostly (1) for P2Y₁₂ inhibitor durations <16 months and/or (2) among patients discharged before July 1, 2011. Therefore, these inconsistent estimates did not impact our interpretation of primary analyses and were less concerning when applying findings to current clinical practice. See Supplemental Tables S8-S13 for sensitivity analyses results.

4. Discussion

In this real-world observational retrospective cohort study, clone-censor-weight marginal structural models emulated a stratified RCT to assess clinical outcomes of 12- to 24-month P2Y₁₂ inhibitor durations following an AMI hospitalization among men and women ≥66 years old. Durations between 16 and 24 months were identified as most likely to meet assumptions for no residual confounding and to therefore be comparable for causal inference in both men and women. Men taking 24-month therapy had the lowest 2-year risk of the composite effectiveness outcome without evidence of increased bleeding risk compared to men taking shorter durations. Women taking 16-month compared to 24-month P2Y₁₂ inhibitor therapy had similar 2-year risks for the composite effectiveness outcome but a 0.7 % lower absolute risk (12 % lower relative risk) of hospitalized bleeds.

The directionality of our composite effectiveness findings was consistent with most of the 7 RCTs that reported results separately in men and women: longer vs. shorter durations appeared beneficial in men while longer vs. shorter durations lacked benefit in women [[6], [7], [8], [9], [10], [11], [12]]. Beyond that, comparisons of our findings to these RCTs are difficult. Our study population was older and had more women, only 71 % underwent PCI with stent placement (100 % in RCTs), and 100 % had an index AMI (0–56 % across RCTs). Additionally, 16- to 24-month P2Y₁₂ inhibitor durations were identified as comparable in our study, but RCTs often compared shorter durations (3- or 6-month vs. 12-month) [6,9,10] or included durations longer than 24 months [8,11,12]. Like our study, composite major adverse cardiovascular and cerebrovascular events in RCTs included death, AMI, and stroke, but differed slightly across trials (e.g. cardiovascular mortality only or composite outcome included bleeding). Finally, cloned patients' data in our study were “assigned” P2Y₁₂ inhibitor durations after surviving 30 days post-AMI, and then 2-year outcomes were assessed; in RCTs, 12- to 24-month outcomes were often assessed following a 6- to 12-month event-free landmark period for study eligibility.

Our results differed substantially from the only other observational study evaluating effectiveness and safety for DAPT durations separately in men and women [21]. While differences in eligibility criteria, measures, design, and analyses explain many differences [19], we believe deficiencies exist in their time-varying Cox proportional hazard models that led to conflicting findings: e.g., bleeding rates doubled with physician-recommended DAPT discontinuation (compared to continuing DAPT) [21]. This is the complete opposite of the expected association: the rate of a known adverse drug reaction should remain stable or decrease if that medication was discontinued. Reverse causality from incorrect temporal ordering of confounders, exposure, and outcome may explain this unexpected finding in their study. Our approach using clone-censor-weight marginal structural models not only encourages diligent consideration for correct temporal ordering of measures [[53], [54], [55],65], but our estimates—analogous to per-protocol RCT estimates—may be easier to interpret than hazard ratios from Cox models with time-varying exposures [59,72].

In our study, >90 % of patients were taking clopidogrel, but current guidelines recommend prasugrel or ticagrelor over clopidogrel after acute coronary syndromes (ACSs) [1,73,74]; however, this recommendation is not without controversy, especially in older patients and patients with higher bleeding risks [75]. Prasugrel is often not recommended in patients >75 years old [1,74,75], and meta-analysis of RCTs in older adults found 1-year ticagrelor versus clopidogrel after ACS had no effect on composite cardiovascular death, AMI, and stroke (RR 1.04; 95 % CI: 0.69, 1.65) but was associated with an increased risk of major or minor bleeding (RR 1.40; 95 % CI: 1.11, 1.76) [76]. Among patients discharged after July 1, 2011 in our study—once all P2Y₁₂ inhibitors were available on the market—4.2 % took ticagrelor, 8.5 % took prasugrel, and 87.3 % took clopidogrel. Our findings are therefore most applicable to patients taking clopidogrel for 12+ months post-AMI, and clopidogrel use remains common in the U.S. Medicare population with data through 2017 [77].

A nationwide registry study in Denmark of patients undergoing PCI for ACS from 2013 to 2016 found that continuing DAPT beyond 12 months versus continuing only aspirin had no benefit in preventing death or major adverse cardiovascular events, but DAPT beyond 12 months came with a 4 % increased absolute risk of hospitalized bleeds over the next 2 years [78]. Comparison to our study is difficult though, as patients in the Danish study were more likely to be younger, male, taking ticagrelor, and to discontinue their P2Y₁₂ inhibitor between 12 and 15 months [19], and analyses were not stratified by sex. However, underpowered sensitivity analyses from the Danish study suggested continuing clopidogrel plus aspirin versus only aspirin beyond 12 months may lower the risk of death and major adverse cardiovascular events [78]. Future research should investigate the effectiveness and safety of durations for specific P2Y₁₂ inhibitor agents in older adults, including strategies of switching from more potent P2Y₁₂ inhibitors to clopidogrel [56].

While we adjusted for baseline diagnosis of atrial fibrillation and baseline use of anticoagulant therapy, we do not recommend using these analyses to guide decisions of using triple versus double therapy [79]. Our analyses did not directly make these comparisons, and patients with an atrial fibrillation diagnosis (~7 %) or using anticoagulant therapy (~7 %) at baseline only made up a small portion of our study sample.

The older men in our study had the greatest benefit in reducing the 2-year risk of death and ischemic outcomes while remaining on P2Y₁₂ inhibitor therapy for 23+ months post-AMI, and this longer P2Y₁₂ inhibitor therapy did not come with a substantial increase in bleeding risk. We hypothesized that we did not observe a clear pattern of increased hospitalized bleeding risk in men attributed to longer P2Y₁₂ inhibitor durations because (1) the effect size was small for this type of bleeding outcome among men primarily taking clopidogrel, or (2) there were heterogeneous treatment effects in men. In exploratory analyses, we stratified patients by the presence/absence of high-bleeding-risk characteristics at the time of index AMI (results not shown). The absolute risk of ischemic and bleeding outcomes both differed greatly between these low- and high-bleed-risk strata, but our findings were consistent when comparing P2Y₁₂ inhibitor durations within the strata separately: longer P2Y₁₂ inhibitors appeared to benefit men by lowering the risk of the composite ischemic outcome without increasing the risk of hospitalized bleeds substantially. Among men primarily taking clopidogrel, the effect size for hospitalized bleeds attributed to longer P2Y₁₂ inhibitor durations investigated within our study may have been too small to precisely detect, but unmeasured confounding and outcome measurement error cannot be completely ruled out.

U.S. clinical guidelines have previously made similar recommendations for P2Y₁₂ inhibitor durations for both men and women following an AMI, including recommendations of treatment for “up to” 12 months for patients managed medically or with fibrinolysis [80,81]; durations of “at least” 12 months for AMI patients managed with PCI and stents, with “up to” 15 months being reasonable with drug-eluting stents [82,83]; or durations of “at least” 12 months for all AMI patients regardless of the intervention strategy [1]. We adjusted for type of AMI and AMI intervention strategy at baseline, but future studies should directly investigate how the type of AMI and the type of AMI intervention strategy (e.g., PCI with drug-eluting stent, medical management, etc.) may modify the effect of P2Y₁₂ inhibitor duration on ischemic and bleeding outcomes, separately in men and women.

Our literature review of the RCTs that compared different P2Y₁₂ inhibitors in men and women separately found a fairly consistent pattern across studies where longer durations tended to benefit men for the reduction of major adverse cardiovascular and cerebrovascular events while women received little to no benefit from longer durations in most of these trials [[6], [7], [8], [9], [10], [11], [12]]. We would argue the clinical guidelines should consider in greater detail how men and women may respond differentially to longer vs. shorter P2Y₁₂ inhibitor durations. Additionally, the guidelines are largely based on evidence from RCTs, but these trials have underrepresented older patients, especially older women [84,85].

In the current study, men taking 23+ months of P2Y₁₂ inhibitor therapy had the lowest 2-year risk of the composite ischemic effectiveness outcome on average, with little to no increased bleeding risk. However, women who continued P2Y₁₂ inhibitor therapy beyond 15–17 months post-AMI had an increased bleeding risk with no benefit in risk reduction of the effectiveness outcome. Our previous findings suggested women should not receive shorter P2Y₁₂ inhibitor durations than comparable men simply because they are female [19], but that final matched sample included only 69 % of the cohort, with the female matched sample becoming more similar to the full male sample.

In this study, given that women continuing P2Y₁₂ inhibitor therapy beyond 15–17 months post-AMI received no benefit from a reduction in the risk of the composite effectiveness outcome but were at a greater risk of harm from hospitalized bleeds, on average, continuing P2Y₁₂ inhibitor therapy beyond 15–17 months post-AMI may be more harmful than beneficial. However, when an older woman's clinical presentation is more similar to that of the male post-AMI Medicare population (e.g., <76 years old and explicit clinical factors suggesting ischemic benefits outweigh bleeding risks), then P2Y₁₂ inhibitor durations longer than 15–17 months post-AMI may be considered.

4.1. Limitations

Unmeasured confounding is a major concern when using observational research for causal inference, and the validity of our estimates assumes no residual time-varying confounding. We adjusted for clinical factors that may impact prescriber recommendations to continue/stop P2Y₁₂ inhibitor therapy (e.g., prescriber-perceived need characteristics in Supplemental Tables S1-S2) and factors that may impact patient adherence to preventive therapy (e.g., enabling, and patient-perceived need characteristics in Supplemental Tables S1-S2). However, residual confounding is still possible from unmeasured variables that may affect both outcomes and decisions regarding duration of P2Y₁₂ inhibitor therapy (e.g., laboratory measures for kidney function). Using negative and positive control outcomes allowed us to identify therapy duration comparisons where this assumption of no residual time-varying confounding most likely held. We focused our interpretation on findings for 16- to 24-month P2Y₁₂ inhibitor durations where we were most confident that this assumption held. However, if readers disagree with our determination of which groups were most comparable for causal inference, we included results for all 12- to 24-month durations for their interpretation.

Second, the validity of our estimates assumes correct model specification. The parameterization of our IPW models were informed from the literature [20,52,[56], [57], [58]], but model misspecification could lead to extreme weights, near violations of positivity, and bias. Our IPWs were truncated to mitigate this bias [65,68]. Additionally, our cumulative incidence model was parametric, encoding when and how discrete time hazards were allowed to vary over time and between comparison groups. A nonparametric model that does not make any assumptions about the functional form of these cumulative incidence curves would have been preferred, but we were not able to use a “saturated” marginal structural model due to the sparsity of outcome events in some follow-up periods (especially hospitalized bleeds in the months following P2Y₁₂ inhibitor discontinuation) [86]. A restricted cubic spline for the assigned months of P2Y₁₂ inhibitor therapy, categorization of follow-up time into the smallest intervals feasible with our data, and an interaction term between these variables (Supplemental Eq. S2) allowed the discrete time hazards some flexibility to vary over time and between comparison groups [65].

Third, we used existing standards for health services research using secondary claims data, but variable misclassification is possible given administrative claims are designed for billing, not research. Our ischemic effectiveness and bleeding safety outcomes were based on ICD-9 diagnoses in the primary position, which typically have higher PPV but lower sensitivity compared to ICD-9 codes in any position [87]. In RCTs and prospective cohorts/registries, bleeding outcomes are often categorized by type and/or severity of the bleed, and comparative safety bleeding outcomes assessed based on the type/severity of bleeding may be preferred to facilitate clinical decision-making regarding the optimal duration of P2Y₁₂ inhibitor therapy in this population [12,88]. However, our data using ICD-9 diagnoses to assess bleeding outcomes does not capture data on hemoglobin drop or hemodynamic instability and cannot classify bleeding outcomes based on these scales [43]. Despite this, ICD-9 code algorithms using the primary diagnosis position have shown high PPV in identifying hospitalized bleeds in populations using anticoagulants or antiplatelets [38,43], and the bleeding outcomes missed by algorithms using primary diagnosis positions only may be more likely to be bleeding that occurred during a hospitalization (as opposed to the bleed being the primary reason the hospitalization occurred) [38]. We would argue that hospitalization specifically for a bleeding event is a clinically distinct and meaningful outcome in this population.

Finally, when using “gap days” to measure treatment duration in prescription claims, measurement error is a substantial concern [28]; however, nearly all sensitivity analyses were consistent when varying 30-day drug-free intervals to shorter and longer intervals, lessening concerns regarding this limitation. Additionally, our data source did not have information about why patients discontinued P2Y₁₂ inhibitor therapy. We did adjust for baseline and time-varying patient-perceived need and prescriber-perceived need characteristics, as well as time-varying prescriber characteristics. Therefore, our models estimate risks attributed to time on therapy but as a weighted average between prescriber-recommended discontinuation and patient-nonadherence discontinuation, assuming no residual confounding attributed to reasons for therapy discontinuation.

5. Conclusions

Among patients ≥66 years old who started a new course of P2Y₁₂ inhibitor therapy (primarily clopidogrel) following an AMI and remained on therapy for at least 12 months, our clone-censor-weight analysis of observational data to emulate a randomized trial found different optimal durations of therapy for men and women. In older men, up to 23+ months of therapy compared to shorter durations resulted in the greatest benefit in reducing the risk of composite death and ischemic outcomes without substantially increasing bleeding risk. In older women, continuing P2Y₁₂ inhibitor therapy beyond 15–17 months post-AMI resulted in an increased bleeding risk without a benefit in the reduction of composite death and ischemic outcomes. Future research should evaluate how the type of AMI and AMI intervention strategy, use of different P2Y₁₂ inhibitor products, and strategies of switching from more potent P2Y₁₂ inhibitors to clopidogrel impact the benefit-risk tradeoff of different P2Y₁₂ inhibitor durations, separately in older men and older women.

Sources of funding

The database infrastructure used for this project was funded by the Pharmacoepidemiology Gillings Innovation Lab (PEGIL) for the Population-Based Evaluation of Drug Benefits and Harms in Older US Adults (GIL200811.0010); the Center for Pharmacoepidemiology, Department of Epidemiology, UNC Gillings School of Global Public Health; the CER Strategic Initiative of UNC's Clinical and Translational Science Award (UL1TR001111); the Cecil G. Sheps Center for Health Services Research, UNC; and the UNC School of Medicine. Dr. Hickson received support from the NIH National Heart, Lung, and Blood Institute (NHLBI) (National Research Service Award Training Grant No. 4T32HL007055-43) as a post-doctoral research fellow with the Cardiovascular Disease Epidemiology Program at The University of North Carolina at Chapel Hill and from the American Foundation for Pharmaceutical Education (AFPE) Pre-Doctoral Fellowship in Health Outcome Disparities. At the time of submission, Dr. Hickson was supported as a Postdoctoral Fellow in Advanced Geriatrics with the Geriatric Research, Education, and Clinical Center at the Veterans Affairs Healthcare System, Pittsburgh, PA. At the time of acceptance, Dr. Hickson was employed by the US Food and Drug Administration (FDA). The contents do not represent the views of the U.S. Department of Veterans Affairs, the FDA, or the United States Government.

Submission declaration

This manuscript is not under review at any other journal. An earlier version of the results from this work were included as part of Ryan Hickson's PhD dissertation at UNC completed in 2020.

CRediT authorship contribution statement

Ryan P. Hickson: Conceptualization, Methodology, Formal analysis, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Anna M. Kucharska-Newton: Conceptualization, Methodology, Writing – review & editing. Jo E. Rodgers: Conceptualization, Methodology, Writing – review & editing. Betsy L. Sleath: Conceptualization, Methodology, Writing – review & editing. Gang Fang: Conceptualization, Methodology, Writing – review & editing, Supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material