Bleeding risk in patients prescribed dual antiplatelet therapy and triple therapy after coronary interventions: the ADAPTT retrospective population-based cohort studies

Methods

Results

We screened 2544 records, identified 322 studies eligible for inclusion and selected a random sample of 70 for initial data extraction. The saturation criterion (no further new factors identified in 10 consecutive studies) was reached after data extraction from 47 studies (16 RCTs and 31 cohort studies) (Figure 1). We identified 59 potential confounders (seven demography, five medical history, 16 comorbidities, six presentation risk, four risk scores, seven biochemical markers and 14 procedural risk), as shown in Table 1.

FIGURE 1.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram.

Methods

Results

Eight cardiac surgeons, six cardiologists and eight GPs were initially approached. Six interviews with cardiac surgeons, six with cardiologists and five with GPs were organised. The remainder of the clinicians either did not reply to the researcher’s e-mails or declined to participate, citing lack of time as the reason. Five interviews were conducted face to face and the rest were conducted over the telephone. Interviews lasted between 26 and 45 minutes.

Patient-related factors

Clinician-related factors

Drug characteristics

When choosing between agents, agent potency was an important consideration for managing the risk of bleeding for a patient:

The reason I’m choosing apixaban is it’s the anticoagulant which probably has [...] the lowest bleeding risk, so because you’re also giving a patient two other drugs which cause bleeding, I would go for the drug which has the lowest bleeding risk.

Cardiologist 001

In this case, there really only is the clopidogrel that we would use because we would not want to combine a very potent antiplatelet agent such as ticagrelor or prasugrel with anticoagulation.

Cardiologist 014

Ticagrelor is a very powerful antiplatelet agent, so [...] I would stop the aspirin and see whether, with ticagrelor only, the nose bleeding and the bruising reduced. If it [...] I would stop the ticagrelor and put the patient back on aspirin and clopidogrel.

Cardiac surgeon 015

The licensing of agents for specific clinical scenarios would also influence prescribing:

For any valvular disease, like if you have AF and you have had a mitral valve repair, mitral valve replacement, or aortic valve replacement, you cannot currently use NOACs or apixaban because they are not licensed for it at the moment.

Cardiac surgeon 011

Local contexts

Methods

Two online surveys (one for cardiologists and one for cardiac surgeons) were developed by the study team, including a methodologist with expertise in survey design, a consultant cardiologist and a consultant surgeon. The surveys were designed to do the following:

1. Describe DAPT prescribing practice among various patient subgroups [based on age, type of event (e.g. ACS vs. non-ACS), concomitant anticoagulant use for cardiology patients and type of surgery, anticoagulant use for cardiac surgery patients].

2. Identify the five most important factors that influence the choice of DAPT prescription in each of six separate domains (e.g. demography, comorbidity and procedure related-characteristics) for cardiology patients and two domains (age and comorbidity only) for cardiac surgery patients. The factors included in the survey were those identified from the systematic review (see Table 1) and additional factors identified from the clinician interviews.

3. Identify whether or not the chosen factors influenced prescribing decisions because they increased risk of ischaemia, risk of bleeding or risk of both ischaemia and bleeding.

The factors and their respective domains were those identified from the systematic review and clinician interviews. The surveys were uploaded to SurveyMonkey^® (Palo Alto, CA, USA) and the online surveys were piloted among a small group of cardiologists and cardiac surgeons to ensure ease of use and to test face and content validity. An invitation including a link to the survey was disseminated by e-mail via the Society for Cardiothoracic Surgery (cardiac surgeons) and the British Cardiovascular Intervention Society (cardiologists) to all individual fellows and members of the societies. The data analysis tools in SurveyMonkey and Microsoft Excel^® (Microsoft Corporation, Redmond,WA, USA) were used to calculate descriptive statistics.

Characteristics of survey respondents (cardiologists and cardiac surgeons)

There were 101 cardiologists and 36 cardiac surgeons who initiated the survey. Of these, 22 cardiologists (22%) and five cardiac surgeons (14%) consented to participate (selected the ‘agree’ electronic consent button) but did not complete any survey questions, so they were removed, leaving a total of 79 cardiologist and 31 cardiac surgeon respondents (Table 4). Of the cardiologists, almost two-thirds of respondents were consultant grade and were evenly distributed across years of practice categories, whereas, for cardiac surgeons, the majority of respondents (90%) were consultants and over half of them had practised for > 15 years. Respondents represented all regions of the UK, but the regions most represented were London, the south west and the north west. Just under two-thirds of cardiologist respondents prescribed DAPT daily, and the majority of the remainder prescribed DAPT two or three times per week. The most common guidelines used by both clinician groups were NICE1 and European Society of Cardiology2 guidelines, although just under half of cardiac surgeon respondents (42%) reported using none of the guidelines. Most cardiologists reported that local protocols for DAPT prescribing were available, whereas two-thirds of cardiac surgeons reported that they had no local protocols for antiplatelet prescribing.

Survey results: cardiologists

Dual antiplatelet therapy prescribing practice for ACS and stable angina patients is shown in Table 5. The default prescribing regimen for ACS STEMI patients was AT (more than two-thirds of respondents, with most prescribing for 12 months). Relatively few STEMI patients were prescribed AP (12% and 5% in those aged ≤ 75 years and > 75 years, respectively).

Among patients who had a non-ST elevation myocardial infarction (NSTEMI), just over half of all respondents prescribed AT for both younger and older patients, whereas, for conservatively managed ACS patients, 41% (for patients aged ≤ 75 years) and 30% (for patients aged > 75 years) of respondents prescribed AT. The use of AP was infrequent, except for STEMI patients aged ≤ 75 years, for whom just over 12% of respondents prescribed this regimen.

Across all ACS groups, approximately 10% of respondents prescribed DAPT for 6 months only (all regimens and age groups), although there was variation in the duration of DAPT treatment in the ACS conservatively managed patient group, with DAPT prescribing ranging from 3 months to > 12 months. Among patients with stable angina undergoing PCI, the default DAPT regimen was AC (for ≈ 90% of patients across both age groups), with variation in duration of treatment ranging from 1 month to > 12 months. Fewer than 10% of stable angina patients were prescribed AT.

Antiplatelet prescribing practice for ACS patients who also need anticoagulants is shown in Table 6. The majority of respondents (63–70%) prescribe TT with AC to ACS patients undergoing PCI (STEMI, NSTEMI and unstable angina patients), with the duration of TT ranging from 1 to 6 months, although 1 month was most frequent. About one-third of respondents prescribe antiplatelet monotherapy, with the majority (80%) prescribing it for 12 months. Most respondents (84%) reported stopping aspirin when stepping down from TT to dual therapy; only 16% reported stopping the P2Y₁₂ inhibitor.

For patients with conservatively treated ACS, antiplatelet monotherapy was most commonly prescribed (62% of respondents), followed by TT with AC (34% of respondents), mostly prescribed for 1–3 months. None of these patients was prescribed AP.

The patient factors that cardiologists take into account when prescribing DAPT are shown in Figure 2a. Patient age, use of concomitant anticoagulation, and whether or not a patient had a current, or had experienced a previous, bleed were considered important by 73%, 72% and 70% of respondents, respectively, followed by disease complexity (40%), anaemia (37%) and renal impairment (33%). About one-quarter considered peptic ulcer diseases, body mass index (BMI), diabetes and ACS risk score as important when prescribing DAPT for their patients.

FIGURE 2.

Patient, procedure- and presentation-related factors and blood test results that cardiologists consider important when prescribing DAPT to their patients. (a) Patient factors; (b) presentation-related factors; (c) blood test results; and (d) procedure-related factors. BMS, bare-metal stent; BVS, bioabsorbable vascular scaffold; CKD, chronic kidney disease; CRUSADE, Can rapid Risk stratification of Unstable angina patients suppress ADverse outcomes with Early implementation of the American College of Cardiology/American Heart Association guideline; DES, drug-eluting stent; ECG, electrocardiogram; GFR, glomerular filtration rate; GRACE, Global Registry of Acute Coronary Events; HAS-BLED, hypertension, abnormal liver/renal function, stroke history, bleeding history or predisposition, labile international normalised ratio, elderly, drug/alcohol usage; HbA_1c, glycated haemoglobin; IABP, intra-aortic balloon pump; ISR, in-stent restenosis; LV, left ventricular; ST, stent thrombosis; SYNTAX, SYNergy between PCI with TAXus and cardiac surgery; TIMI, thrombolysis in myocardial infarction.

Presentation-related factors and blood test results that cardiologists consider when they prescribe DAPT are shown in Figure 2b and c, respectively. Presenting syndrome (ACS or stable angina) was the single most important presentation-related factor taken into consideration when prescribing DAPT (98% of all respondents). In terms of blood test results, haemoglobin (82%), platelet count (69%), glomerular filtration rate (GFR) (54%), creatinine (49%) and troponin (47.5%) were considered to be important.

Procedure-related factors that influence DAPT prescription are shown in Figure 2d. More than 70% of respondents thought that stent failure (76%), multivessel PCI (76%) and length of stented segment (71%) were important factors that influenced their DAPT prescription; 55% thought that type of stent used was important when prescribing DAPT. Just over one-third (37%) considered infarct-related characteristics, and about one-quarter considered coronary complications or whether a native or graft vessel had been stented.

Respondents were asked to indicate whether or not the factors that influence their decision-making process when prescribing DAPT did so because of their association with ischaemia risk, bleeding risk, or both ischaemia and bleeding risk. Figure 3 shows the contribution of each risk (ischaemia, bleeding or both) to the decision-making process for the most commonly considered factors (i.e. those regarded as important by ≥ 50% of respondents). Figure 2 indicates that cardiologists consider bleeding and ischaemia risks equally when prescribing DAPT. Five of the 12 factors (presenting syndrome, ACS or stable angina, and those related to the PCI procedure) were chosenmainly on the basis that they increase ischaemia risk, 5 of the 12 (concomitant anticoagulation, previous or current bleeding, bleeding risk score, haemoglobin level and platelet count) were chosen mainly on the basis that they increase bleeding risk, and 2 of the 12 (age and GFR) were chosen mainly on the basis that they increase both bleeding and ischaemia risks.

FIGURE 3.

The contribution of each risk (ischaemia, bleeding or both) to the cardiologist decision-making process for the most commonly considered patient, presentation- and procedure-related factors (i.e. those regarded as important by ≥ 50% of respondents).

Survey results: cardiac surgeons

Table 7 shows antiplatelet prescribing for the various patient subgroups. Antiplatelets were commonly prescribed for the following patient subgroups: CABG and recent ACS (81%), CABG with poor vessel/conduit quality (71%) and CABG and previous stent (61%). Relatively few respondents prescribed DAPT as a substitute for vitamin K antagonist prophylaxis for CABG and tissue valve surgery (13%) and post-operative AF (3%).

For the CABG with recent ACS, CABG with poor vessel/conduit quality and CABG with previous stent patient subgroups, DAPT was the preferred treatment (76%, 79% and 86%, respectively). DAPT with clopidogrel was most frequently prescribed for the last two patient subgroups, although, for the CABG and recent ACS patient subgroup, respondents were equally likely to prescribe DAPT with ticagrelor. Low-dose aspirin was preferred for patients undergoing off-pump CABG (62% of respondents), patients undergoing CABG and valve surgery (76%) and patients with post-operative AF (76%). Just under two-thirds of respondents (18/29; 62%) never prescribe DAPT after surgery for patients who require thromboprophylaxis with warfarin or a NOAC, whereas the remainder (38%) prescribe DAPT only in very selected patients requiring warfarin or a NOAC.

The patient factors that cardiac surgeons take into account when prescribing DAPT are shown in Figure 4. Previous or current bleeding and previous CABG or PCI were considered important by 59% and 66% of respondents, respectively, followed by age, concomitant anticoagulation and bleeding risk score (48%). About one-third considered previous stroke and peptic ulcer disease, one-quarter considered disease complexity and one-fifth considered resistance to antiplatelet agents to be important.

FIGURE 4.

Patient factors that cardiac surgeons consider important when prescribing antiplatelet agents to their patients. CKD, chronic kidney disease; CRUSADE, Can rapid Risk stratification of Unstable angina patients suppress ADverse outcomes with Early implementation of the American College of Cardiology/American Heart Association guideline; GRACE, Global Registry of Acute Coronary Events; HAS-BLED, hypertension, abnormal liver/renal function, stroke history, bleeding history or predisposition, labile international normalised ratio, elderly, drug/alcohol use; SYNTAX, SYNergy between PCI with TAXus and cardiac surgery; TIMI, thrombolysis in myocardial infarction.

Respondents were asked to indicate whether or not the factors that influence their decision-making process when prescribing DAPT did so because of their association with ischaemia risk, bleeding risk, or both ischaemia and bleeding risk. Figure 5 shows the contribution of each risk (ischaemia, bleeding or both) to the decision-making process for all patient factors. Both bleeding and ischaemia risks were considered equally in the decision-making process, but concern about bleeding risk featured more prominently in the factors that > 48% of surgeons considered to be important when prescribing (previous or current bleeding, age, concomitant anticoagulation and bleeding risk score).

FIGURE 5.

The contribution of each risk (ischaemia, bleeding or both) to the cardiac surgeon decision-making process for the most commonly considered factors (i.e. those regarded as important by ≥50% of respondents). CKD, chronic kidney disease; GRACE, Global Registry of Acute Coronary Events; TIMI, thrombolysis in myocardial infarction.

Data sources

The CPRD is a database of primary care electronic health record data (available online via CPRD GOLD) from participating general practices, covering 7% of the UK population. 26 Patients included in CPRD are largely representative of the UK population in terms of age, sex, ethnicity and BMI. HES cover all hospital admissions for all English patients whose treatment is funded by the NHS, whether treated by the NHS or by independent providers. 27 Seventy-five per cent of English general practices included in CPRD are linked to HES data. 26 We obtained data from 1 April 2009 to 31 July 2017; this period covers the introduction of the newer antiplatelet agents prasugrel and ticagrelor. The study protocol was approved by the Independent Scientific Advisory Committee of the CPRD (protocol number 16_126R).

Study populations

We initially specified three target trials for (1) patients undergoing CABG, (2) patients hospitalised and conservatively managed for ACS and (3) patients undergoing PCI. Eligibility and exclusion criteria for the three target trials are listed in Table 9. However, for the purpose of the statistical analysis, patients undergoing PCI were separated into emergency PCI and stable PCI, as these represent different populations: patients undergoing emergency PCI have ACS, which is associated with poorer short-term prognosis than PCI for stable coronary artery disease. Some analyses in the emergency PCI population were restricted to the STEMI population only.

Patients were included if they had a PCI, CABG or ACS diagnosis (index event) recorded in HES during the study period (1 April 2010 to 31 January 2017) and had at least 1 year of linked CPRD–HES data before the date of their index event. They must also have been prescribed one of the treatment regimens being compared in the target trial corresponding to their index event. One year’s data preceding eligibility for the target trial is adequate to apply most of the exclusion criteria and determine comorbidities and medication history; such information would be collected at baseline in a randomised trial. The following Office of Population Censuses and Surveys (OPCS) procedure codes (PCI and CABG) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), codes (ACS no procedure) were used to identify patients: PCI: K49, K50 and K75; CABG: K40, K41, K42, K43, K44, K45 and K46; ACS without a procedure: I20.0, I21, I22, I24.9 with no OPCS code for PCI or CABG in the same hospital admission; PCI STEMI: K49, K50 and K75 and I21.0-I21.3 or I22.0–I22.8 as primary or secondary diagnosis.

Interventions

The interventions of interest for the three target trials are shown in Table 9. Guidelines recommend low-dose aspirin (75–100 mg per day) plus either clopidogrel (75 mg per day), prasugrel (5 mg or 10 mg per day) or ticagrelor (90 mg twice per day) for PCI and conservatively managed ACS patients. For PCI patients, the interventions of interest were AC, AP and AT. In conservatively managed ACS patients, clopidogrel is the most commonly prescribed second antiplatelet agent (in addition to aspirin), and a large proportion of patients are prescribed aspirin only; therefore, the interventions of interest are aspirin only (any dose) and AC. There is variation in aspirin prescription for CABG patients (75–300 mg per day). Surgeons may also prescribe an additional antiplatelet agent, most commonly clopidogrel. Therefore, the comparisons of interest for CABG patients are aspirin only (any dose, reflecting variations in usual care in different hospitals) and AC (75 mg per day).We specified these comparisons based on preliminary feasibility counts from the CPRD, which showed that few CABG and conservatively managed ACS patients are prescribed AP or AT. Product codes for the antiplatelets are detailed in Appendix 5.

In the target trials, the interventions would be assigned during the hospital stay, as soon as patients were eligible for antiplatelet therapy. Our observational data set does not have information on medication given to patients at discharge because HES does not include medications data. Therefore, the first time at which we have information on the antiplatelet regimen to which patients were assigned in hospital is when they receive their first primary care prescription(s) for aspirin or DAPT, recorded in the CPRD.We used the first prescription in the CPRD as a proxy for the medications that patients started in hospital. This is justified because the qualitative study with clinicians (see Chapter 2) supported the assumption that GPs are unlikely to change the prescriptions that were started in hospital.

We classified patients according to the first prescription recorded in the CPRD in the first 2 months after hospitalisation for PCI, CABG or ACS. This 2-month window was based on variability in the amount of DAPT medication provided to patients in hospital following their PCI, CABG or ACS treatment, and hence variability in the time when they first requested a repeat prescription from their general practice. A preliminary investigation showed that > 75% of eligible patients had a prescription for one or more antiplatelet agents during this time period. Patients were assigned to intervention groups as follows: (1) if a patient had a prescription only for aspirin during the 2-month window after hospital discharge, they were assigned to an aspirin-only intervention; (2) if a patient also received a prescription for clopidogrel, prasugrel or ticagrelor, they were assigned to AC, AP or AT; and (3) if there was a prescription for more than one additional antiplatelet agent in the 2-month window, the patient was assigned to an intervention based on the agent prescribed first.

For example, if a patient had an aspirin prescription and a prescription for clopidogrel before a prescription for ticagrelor, the patient was assigned to the AC intervention. Patients with no prescriptions in the CPRD for aspirin or AC, AP or AT within the 2-month window were excluded from the main analysis. Patients who experienced a major bleed or a MACE prior to the first antiplatelet prescription(s) occurring in the CPRD within 2 months of the index event were also excluded from the analysis because we could not assume that the antiplatelet prescription observed in the CPRD would be the same as that assigned in hospital at the time of the index event. Both these groups of patients (those with no antiplatelet prescriptions and those who experienced an event) were included in a sensitivity analysis in which assignment to DAPT was performed using multiple imputation for missing data based on propensity scores; see Sensitivity analysis 1: multiple imputation for unknown intervention group.

Outcomes

The primary outcome was any bleeding event. For each patient, we identified all bleeding events in HES and the CPRD during follow-up (365 days after the index event). We originally planned to classify bleeding events according to the Bleeding Academic Research Consortium (BARC) bleeding scale;28 however, the data sets did not contain all of the information required to allow BARC classification. We have specified a comprehensive list of bleeding codes in the CPRD and HES (see Appendix 6). These were categorised according to anatomical site for descriptive purposes. Secondary outcomes were as follows: any major bleeding event, any minor bleeding event, all-cause mortality, cardiovascular mortality, mortality from bleeding (these would be identified from linked Office for National Statistics data), MI, stroke, additional coronary intervention and MACEs (defined as any of MI, stroke, cardiovascular mortality or additional coronary intervention). Major bleeding events were defined as any HES bleed, and minor bleeding events were defined as any CPRD bleed without any HES bleed being recorded within 28 days (i.e. ± 14 days of the CPRD bleeding event). This was to ensure that the record in the CPRD is not a duplicate count of a HES bleeding event, as the CPRD records GP-reported bleeds and may also record bleeds leading to hospital admission.

Follow-up

The start of follow-up (the index event) was the date of the index hospital procedure (PCI and CABG groups) or the start date of the hospital episode that contained the ACS diagnosis (ACS group). Patients were followed up for 365 days after the index event.

Confounding and co-interventions

Confounders (variables that predict both risk of bleeding and intervention group) were specified a priori29,30 (see Chapter 2). Read, ICD-10 and product code lists were prepared to identify all confounders, either from published sources31 or created by the study team [methodologists familiar with Read (CPRD) and ICD-10 (HES) coding systems and clinicians]. Code lists for the confounders are shown in Appendix 7.

Sample size

The estimated rates of any bleeding with the different therapies are 5% for aspirin, 9% for AC and 12% for AP and AT. 6,7,32,33 We used preliminary feasibility counts provided by the CPRD to identify numbers of patients eligible for each target trial: (1) PCI: AC (reference: 6738 patients) versus AP (842 patients) or AT (770 patients), (2) CABG: aspirin (reference: 2556 patients) versus AC (595 patients) and (3) conservatively managed ACS: aspirin (reference: 8148 patients) versus AC (3082 patients). These estimates gave expected numbers of bleeding events of at least 700 for PCI, 180 for CABG and 680 for ACS, assuming ratios of 8 : 1 (AC : AP or AC : AT) for PCI, 4 : 1 (aspirin : AC) for CABG and 2.5 : 1 (aspirin : AC) for ACS. The HRs detectable with 90% and 80% power at 5% statistical significance, assuming the group ratios given previously, are shown in Table 10. The correlation of the DAPT with other covariates adjusted for was unknown and we assessed the impact of a range of correlations (0, 0.3 and 0.5).

Statistical analyses

We examined temporal changes in DAPT prescribing and bleeding for PCI, CABG and ACS populations between 2010 and 2017. Descriptive statistics were used to summarise the characteristics of the different intervention groups and standardised mean differences (SMDs) were used to compare them. We estimated the rates of any bleeding (number of events/person-time) with 95% confidence intervals (CIs) for each group. We separated major (leading to hospital admission, i.e. HES inpatient data) and minor (CPRD) bleeding because adverse events of each type have different health and resource use consequences. We censored all bleeds at the GP transfer-out date or last collection date, thereby ignoring any bleeds in the HES data set recorded after this period.

Data-cleaning and dealing with missing data

The most recent record of smoking status and BMI were used; data-cleaning rules suggested by Atkinson et al. 34 were used for smoking and rules suggested by Bhaskaran et al. 35 were used for BMI. All data, such as prescription dates, recorded after the date of death were set to missing. For binary variables, we took no record of a code in the data sets to mean absence of event. We examined all non-binary variables for missing data. Smoking and BMI were missing 4% and 8% of values, respectively, for the emergency PCI group; 1% and 5%, respectively, for the CABG group; and 4% and 7%, respectively, for the ACS group. These missing data were replaced with age- and sex-adjusted averages.

Comparative analysis

Analyses estimated the effects of assigned intervention (analogous to an intention-to-treat analysis of a randomised trial) for the antiplatelet regimens corresponding to the first prescription of aspirin or DAPT in the CPRD (see Interventions). For each target trial, we calculated propensity scores for the comparative antiplatelet regimens using a backward stepwise logistic regression with significance level for removal from the model set at 0.25. For the emergency PCI target trial (2012–17), we report only results pertaining to AT versus AC, as AP is almost exclusively prescribed for STEMI patients. The AC versus AP versus AT analysis was performed in the STEMI population only, using a multinomial logistic regression to accommodate the three interventions (AC, AP and AT).

All confounders identified as possibly being related to the bleeding outcome were included in these stepwise models. Criteria for excluding tails of propensity score distributions were decided by reviewing the bleeding events between interventions, based on cut-off points of the propensity score at fifth, 25th, 50th, 75th and 95th percentiles. We excluded subjects in the extreme tails of the propensity score distribution (lower fifth percentile of propensity scores) only for CABG analyses when few patients or deaths in either intervention were observed. There was good overlap of covariate distributions for emergency PCI, STEMI PCI and conservatively managed ACS. All subsequent analyses were based on data with these tails excluded (CABG analyses only). This made it more likely that analyses were restricted to patients eligible to receive either intervention. Kaplan–Meier curves were generated after adjusting by the inverse probability of treatment weights using the propensity scores,36 where the weights were defined as 1/propensity score for the treatment received.

We used Cox regression models to estimate crude and adjusted HRs with 95% CIs for the time to first bleeding event, comparing intervention groups for each target trial. Participants free from a bleeding event were censored at 12 months after the index event. For each target trial, we adjusted for all potential confounders identified in Chapter 2 and the propensity score. 37,38 All continuous variables (calendar year, age, BMI and propensity scores) were included in models as cubic splines with knots set at the 25th and 75th percentiles. Visual assessments of these splines were undertaken to check that these were appropriate. Confounders were included using a backward stepwise approach with significance level for removal from the model set at 0.25, and additionally adjusted for propensity scores. The intervention group was included in all models. We could not formally compare interventions among the stable PCI patients because there was no variability in treatment: > 93% were prescribed AC.

For all secondary end points, we used survival models to estimate adjusted HRs with 95% CIs for time to first event, as detailed previously. For mortality outcomes, we adjusted for a smaller list of confounders (year, age, sex, BMI, ethnic group, smoking, Charlson Comorbidity Index score and propensity scores). 39 The Charlson Comorbidity Index score was calculated separately using Read codes and ICD-10 codes for the year prior to the index event. The Read codes were extracted from Khan et al. 40 and the ICD-10 codes were extracted from Maringe et al. 41 For both the Read codes and the ICD-10 codes, each diagnosis was considered only once and the Charlson Comorbidity Index score was calculated by adding the scores associated with these diagnoses. The final Charlson Comorbidity Index score was taken as the higher value calculated via Read codes and ICD-10 codes for each patient. In the very few instances [20 (0.2%) emergency PCI patients and three (0.1%) ACS patients] for which there were no GP data or HES data in the preceding year, the Charlson Comorbidity Index score was set to zero.

Sensitivity analyses

Subgroup analyses

For each subgroup analysis, the main primary outcome analysis (adjusted by propensity scores and all selected confounders) was repeated including an interaction term for the subgroup. The following subgroups were investigated: ACS versus non-ACS (CABG population), defined by the presence or absence of a diagnosis of ACS during the same continuous inpatient spells as the CABG procedure; diabetes versus non-diabetes, defined by the presence or absence of a diagnosis of diabetes in the year prior to the index procedure/event; chronic kidney disease versus non-chronic kidney disease, defined by the presence or absence of a diagnosis of kidney disease in the year prior to the index procedure/event; concurrent prescription for proton pump inhibitors (PPIs), defined by the presence or absence of a prescription for PPIs in the window between discharge after the index event and 2 months (61 days) later. The codes used to identify subgroups were the same as in the confounder coding process.

Treatment switches and adherence

Treatment switch/discontinuation was defined as starting a second antiplatelet or stopping aspirin in the aspirin group or stopping clopidogrel or aspirin in the AC group. Starting a second antiplatelet was defined as a patient receiving at least one prescription for a second antiplatelet during follow-up. Stopping clopidogrel or aspirin was defined as a gap between repeat prescriptions of > 1.5 times the number of days’ supply of the last prescription.

Adherence was defined using the medication possession ratio (MPR). 46 The MPR was calculated as the total number of days of available medication (quantity of drug prescribed divided by the daily dose) divided by 12 (12 months of follow-up). For the AC group, the overall MPR was calculated as the average MPR of AC. Non-adherence was defined as a MPR of < 80%. 47

Statistical analyses were undertaken using Stata^® 15.1 (StataCorp LP, College Station, TX, USA).

Trends in antiplatelet prescribing and bleeding over time

Figure 7 shows the trends in prescription of antiplatelet therapy between 2010 and 2017. Prescriptions with aspirin monotherapy decreased (from 70.9% in 2010 to 52.2% in 2017), whereas prescriptions of DAPT with clopidogrel increased (from 13.2% in 2010 to 29.0% in 2017). Prescriptions of P2Y₁₂ monotherapy were stable over time (average 3.5%) and a proportion of patients (average 12.9%) received no antiplatelet therapy. Rates of bleeding did not change markedly over time; on average, the rates of major and minor bleeds were 18.0 and 38.3 events, respectively, per 1000 person-years.

FIGURE 7.

Proportion of CABG patients being prescribed different antiplatelet regimes by year and rates of major (HES), minor (CPRD) and total bleeding events by year among CABG patients. (a) Different antiplatelet regimes; and (b) bleeding events.

The total number of bleeds (major and minor bleeds combined) appeared to increase slightly over time, consistent with the increase in the proportion of patients receiving AC.

Figure 8 shows how the CABG target trial was constructed from the available data. There were 5335 CABG patients in the linked CPRD–HES data set, and 2783 (52%) of them were eligible for inclusion in the target trial (Table 11). Of these, 482 (17%) were excluded because they had no antiplatelet prescription data in the first 2 months after hospital discharge, leaving 2301 (83%) included in the primary analysis.

FIGURE 8.

Flow diagram describing the construction of the CABG target trial. C, clopidogrel; P, prasugrel; SA, sensitivity analysis; T, ticagrelor.

Baseline characteristics of participants included in and participants excluded from the target trial

The baseline characteristics of participants in the CABG target trial are shown in Table 11. We used the SMD to express the size of the difference between groups relative to the variability observed for each patient characteristic. A SMD of 0.10 is the threshold used to denote a meaningful imbalance in the covariates between groups. 48,49 The number of eligible CABG patients decreased every year between 2010 and 2017 because of the decline in the number of practices in the CPRD GOLD over time (Nafiu Ismail, Keele University, 2019, personal communication). 50 For aspirin versus AC, there was fairly good balance in the covariates between groups. Relatively few covariates had SMDs of > 0.10: age (0.13, lower in AC), sex (0.11, more women received AC), ethnic group (0.22, higher proportion received AC among those other than white), history of MI (0.27, higher proportion in the AC group) and clotting disorder (0.14, higher proportion in the aspirin group). There was no difference of in length of hospital stay between the aspirin and the AC groups [median 6, interquartile range (IQR) 5–9, and median 6, IQR 5–8, respectively].

We also used SMDs to identify potential differences between the group of patients for whom the intervention was unknown (n = 482) and the group for whom an intervention could be assigned (aspirin or AC, n = 2301). The unknown intervention group had lower proportions of patients with a history of ischaemic heart disease (IHD) (96% vs. 99%) and hypercholesterolaemia (42% vs. 48%), and higher proportions of patients with heart failure (15% vs. 12%), patients who had had previous surgery (6% vs. 4%) and patients taking steroids (9% vs. 6%) (all SMDs > 0.10). Of the 482 patients without an intervention, 107 (22%) either died or had a major bleed or ACS event before their first prescription and 375 (78%) had no prescription of an antiplatelet agent in the first 2 months after their index event.

The characteristics of patients who died or had a major bleed or ACS event, compared with those who had no antiplatelet prescription within 2 months of discharge, are shown in Table 12. The group of patients who experienced an event included older patients (aged 72 years vs. 66 years, SMD 0.6); more women (23% vs. 19%, SMD 0.12); more ex-smokers (49% vs. 41%, SMD 0.16); and more patients with a history of MI (50% vs. 33%, SMD 0.34), hypertension (70% vs. 63%, SMD 0.15), peripheral vascular disease (17% vs. 11%, SMD 0.18), stroke (3% vs. 1%, SMD 0.13), heart failure (26% vs. 12%, SMD 0.36), renal disease (11% vs. 4%, SMD 0.27), cancer (12% vs. 6%, SMD 0.21) and anaemia (7% vs. 3%, SMD 0.17); and fewer patients with hypercholesterolaemia (37% vs. 44%, SMD 0.13). Furthermore, more patients who experienced an event were taking steroids (15% vs. 8%, SMD 0.23).

Bleeding events among participants included in and those excluded from the target trial

Of the 2186 patients included in the target trial, 111 (5%) experienced at least one bleeding event: 69/1596 (4%) in aspirin and 42/590 (7%) in the AC group. With regards to major and minor bleeding events, 38/2186 (2%) patients experienced a major bleed and 79/2186 (4%) experienced a minor bleed. The proportion of patients experiencing a major and minor bleeding event in aspirin were 20/1596 (1%) and 53/1596 (3%), respectively, while in the AC group the proportion of patients experiencing a major and minor bleeding event were 18/590 (3%) and 26/590 (4%), respectively.

Figure 9 shows the Kaplan–Meier curves of cumulative bleeding in AC versus aspirin groups [any bleed, major (HES reported) and minor (CPRD reported)], and Table 13 shows the follow-up time and the number of bleeding events for major and minor bleeding. The cumulative incidence of any bleeding was higher with AC than with aspirin (see Figure 9a). The curves diverged early (approximately 1 month) after the index event. This was also reflected in the Kaplan–Meier curve for minor bleeding events (see Figure 9c), whereas, for major bleeding events, the curves diverged after approximately 3 months (see Figure 9b). The crude incidence rate of major bleeds in the AC group was more than double that of the aspirin group (30.9 vs. 12.6 events per 1000 person-years, respectively). The crude incidence rate of minor bleeds was also higher in the AC group than in the aspirin group (45.3 vs. 33.8 events per 1000 person-years, respectively) (see Table 13). Of the 111 (5%) patients who experienced bleeding events, the majority (n = 88, 79%) experienced a single bleed and 23 (21%) experienced more than one bleed. Over half of all bleeds were gastrointestinal in origin; just over one-quarter were ear, nose and throat; and just over 10% were skin and soft-tissue bleeds (Table 14). Bleed sites did not differ markedly between the aspirin and the AC groups, with the exception of ear, nose and throat bleeds, which were more prevalent in the AC group.

FIGURE 9.

Kaplan–Meier curves displaying cumulative bleeding according to intervention group. (a) Any bleeding; (b) major bleeding; and (c) minor bleeding. Plots are weighted according to the inverse probability of treatment received, and so compare outcomes if all eligible patients received aspirin or AC.

Patients who could not be assigned an intervention either because they experienced a bleeding or ischaemic event or died before their first prescription, or because they had no prescription in CPRD within 2 months of discharge (17% of the eligible population) had a slightly higher bleeding rate than the patients included in the target trial (7% vs. 5%) (see Table 15). The bleeding rate was markedly higher in those who experienced an ischaemic or bleeding event or died than in those who had no prescription in CPRD within 2 months of discharge (15% vs. 5%).

Analyses for the primary outcome (bleeding)

The primary analysis excluded patients for whom we could not assign an intervention (n = 482) and those in the lowest fifth percentile of propensity score (n = 115). The patients who could not be assigned an intervention had a higher rate of any bleeding and major bleeding than those included in the target trial (any bleeding: 7% vs. 5%, respectively; major bleeding: 5% vs. 4%, respectively) (Table 15). The crude HR indicated an increase in the hazard of bleeding in the AC group, compared with the aspirin group (1.69, 95% CI 1.15 to 2.48) (see Table 15). The HR was similar after adjustment for propensity scores and confounders (1.72, 95% CI 1.15 to 2.57). When separated by major (HES-reported) and minor (CPRD-reported) bleeding, there was an increased hazard of major bleeding (adjusted HR 2.89, 95% CI 1.48 to 5.64), but not of minor bleeding (adjusted HR 1.22, 95% CI 0.74 to 1.99), in the AC group, compared with the aspirin group.

Bleeding events among patients eligible for the target trial but not included in the primary analysis

Table 15 shows the number of bleeds among patients who were not included in the analysis. Patients who experienced an event or died before their first prescription in the CPRD had a higher rate of major bleeding than those included in the target trial (15% vs. 5%, respectively). Those with no prescription in the CPRD within 2 months of discharge had the same bleeding rate as those included in the target trial (5% vs. 5%).

Mortality and ischaemic events among participants included in and those excluded from the target trial

Figure 10 shows the Kaplan–Meier curves for the secondary outcomes of all-cause and cardiovascular mortality, mortality from bleeding, MI, stroke, additional coronary intervention and the composite outcome of MACE. The event rates for all secondary outcomes were higher among patients excluded from the target trial (15% of the eligible CABG population) (Table 16). However, this event rate was driven entirely by the 107 patients (3% of the eligible population) who died or experienced another ACS event, rather than those with no prescription in the CPRD within the 2-month time window (the latter had an incidence of all secondary outcomes similar to that of patients included in the target trial).

FIGURE 10.

Kaplan–Meier curves displaying cumulative all-cause and cardiovascular mortality, mortality from bleeding and ischaemic events (MI, stroke, additional PCI) and cumulative MACEs, according to intervention group. (a) All-cause mortality; (b) cardiovascular mortality; (c) mortality from bleeding; (d) MI; (e) stroke; (f) additional PCI; and (g) MACE.

Analyses for the secondary outcomes (mortality and ischaemic events)

With regard to ischaemic outcomes, AC increased the hazards of MI (HR 2.45, 95% CI 1.10 to 5.45), additional coronary intervention (HR 2.48, 95% CI 1.13 to 5.46) and MACEs (HR 1.95, 95% CI 1.17 to 3.26). AC did not increase the hazard of mortality or of cardiovascular mortality (see Table 16).

Treatment switches and adherence

Treatment switches in the aspirin and AC groups by type of switch and whether the switch occurred before or after bleeding or an ischaemic event are shown in Table 17.

In the 12 months after the index event, 341 out of 1702 (20%) patients in the aspirin group were identified as ‘switchers’. There were 356 treatment switches; 281 (79%) were aspirin discontinuations, and 75 (21%) initiated a different P2Y₁₂ inhibitor. The median time to switching was > 6 months in the two groups of switchers. On average, patients who initiated a second antiplatelet received 5.9 prescriptions of the second antiplatelet (6 months’ supply). The median time to switching was between 6 and 8 months in all groups of switchers.

Among patients assigned AC, 106 out of 599 (18%) were identified as switchers. There were 151 treatment switches; 85 (56%) were aspirin discontinuations, 41 (27%) were clopidogrel discontinuations, 22 (15%) were aspirin and clopidogrel discontinuations and < 5 were initiations of a different P2Y₁₂ inhibitor. The median time to switching was ≥ 6 months among all those who switched.

Across both groups (aspirin and AC), 42 switchers had a bleed or ischaemic events, 29 (69%) in aspirin and 13 (31%) in AC. Most of these events occurred before the switch.

In the aspirin group, the majority of switchers (> 80%) did not experience a bleeding or ischaemic event; 6% experienced a bleeding event (most before the switch and distributed equally between the two groups of switchers) and 4% experienced an ischaemic event, with the majority of these in the group of switchers who initiated a second antiplatelet agent before the switch.

In the AC group, just 2% of switchers experienced a bleed, with most of these occurring before the switch and in the group of switchers who discontinued aspirin; 1% experienced an ischaemic event, with most of these also occurring before the switch and also in the group that discontinued aspirin.

Adherence, defined as a MPR of ≥ 0.8, was 70% in the aspirin group and 54% in the AC group.

Discussion

This is the first study using routinely collected data to examine the incidence of bleeding among patients undergoing CABG in the 12 months after their procedure. Only 5% of the study population experienced any bleeding event; this is the same as the incidence for any bleeding reported in a 2018 meta-analysis (including five RCTs and eight observational studies),51 but lower than the incidence of major bleeding (7%) from eight observational studies reported in another meta-analysis. 52 These discrepancies probably arise because of different criteria for reporting bleeding events being used in individual studies and different methods of data collection.

The data suggest that there is underascertainment of minor bleeding (including ‘nuisance’ bleeding) in the CPRD. The incidence of minor bleeding was 4% across the CABG population, which is much lower than the incidence of nuisance bleeding (29–38%) reported in previous studies of patients on antiplatelet medication (in which patients were interviewed about bleeding events). 12-14 This suggests that the vast majority of nuisance bleeding does not prompt patients to go to their GP.

In our population, DAPT with AC, compared with aspirin monotherapy, was associated with an increased hazard of any bleeding (HR 1.72, 95% CI 1.15 to 2.57) and major bleeding (HR 2.89, 95% CI 1.48 to 5.64), but not minor bleeding (HR 1.22, 95% CI 0.74 to 1.99). Recent meta-analyses52,53 suggest that DAPT with AC does not increase the risk of major bleeding, compared with aspirin monotherapy. Agarwal et al. 52 pooled data from five RCTs and seven non-randomised studies [RCTs with, pooled, 16 events/446 participants in the AC group and eight events/445 participants in the aspirin group, relative risk (RR) 1.82, 95% CI 0.78 to 4.25; cohort studies with, pooled, 281 events/4398 participants in the DAPT group and 294 events/4327 participants in the aspirin group, RR 1.10, 95% CI 0.94 to 1.29]. The size and direction of the point estimate for the RCT meta-analysis was similar to our analysis, but the CIs were wide, reflecting the small sample size and low frequency of events.

Solo et al. 53 conducted a network meta-analysis comprising 3745 patients and investigating different antithrombotic regimens (DAPT with clopidogrel and ticagrelor, antiplatelet monotherapy, vitamin K antagonists and rivaroxaban). It showed no increase in major bleeding with DAPT with AC versus aspirin monotherapy (RR 0.85, 95% CI 0.30 to 2.37), although the CIs were wide (reflecting the small sample sizes and low frequency of events of the included trials). None of the published meta-analyses evaluated minor bleeding.

The differences in effect size between our study and the meta-analyses are likely to be a result of differences in design, populations and methods of data collection. We also excluded a group of patients eligible for inclusion (15% of the eligible population) who could not be assigned an intervention because they died, had a major bleed or ACS event or had no prescription in the CPRD in the specified time window for assigning the intervention. This may have introduced selection bias into our study, particularly as these patients had a higher rate of bleeding than the included population.

Our results were robust to the different assumptions tested in the sensitivity analyses (multiple imputation for unknown intervention groups, exclusion of patients with high risk of bleeding and no censoring for bleeding events after the last collection date in the CPRD). HRs were comparable across all analyses. A limitation of the multiple imputation is that it was based on patient characteristics for the cohort of patients with known intervention. However, the subgroup of patients who could not be assigned an intervention (482 patients, 15% of the eligible population) had a higher risk of bleeding, and were, therefore, different from the population with known intervention.

Surprisingly, our study also showed an increased hazard of MI, additional coronary interventions and MACE (composite of MI, coronary reinterventions and death) in the AC group, compared with the aspirin group. The meta-analysis by Agarwal et al. 52 found a decreased risk of all-cause mortality (RR 0.67, 95% CI 0.48 to 0.94) and MACEs (composite of MI, stroke and death, RR 0.84, 95% CI 0.71 to 0.99) in the AC group, compared with the aspirin group, with the effect size similar across RCTs and observational studies. Other meta-analyses, for example Verma et al. ,54 showed a similar effect of DAPT on mortality (RR 0.68, 95% CI 0.43 to 1.08) and MACEs (0.86, 95% CI 0.73 to 1.03) among patients after CABG. The network meta-analysis by Solo et al. 53 showed a similar effect size for AC compared with aspirin for all-cause mortality (RR 0.70, 95% CI 0.11 to 4.50) and MI (RR 0.71, 95% CI 0.26 to 1.96).

The unexpected increase in MACEs that we observed in our study suggests that CABG patients who were prescribed DAPT were higher-risk patients. The baseline covariates show that the AC population was younger, had a higher proportion of individuals who were other than white and a higher rate of previous MI. Although these covariates were adjusted for in the statistical analysis, they reflect a population with presumed increasing complexity of disease who will be more likely to experience secondary events. Furthermore, we did not have data on a substantial proportion of potential confounders, in particular procedure-related characteristics. Our results are, therefore, likely to reflect a certain degree of confounding by indication.

We excluded 15% of the eligible population from the primary analysis because they could not be assigned an intervention. Excluding patients who experienced a major bleed or MACE prior to the first (if any) antiplatelet prescription(s) occurring in CPRD within 2 months of the surgery was necessary because we could not reliably assume that the observed treatment would be the same as the assigned intervention at baseline for those patients. However, this may have induced selection bias, as we excluded a high-risk population. For example, a true higher risk of MACE in the aspirin group may be masked or even reversed by excluding susceptible patients (high-risk patients who experienced a major bleeding event or MACE prior their first prescription in primary care) from the risk set. It is well established that depletion of susceptible populations from an analysis population, for example by including prevalent users, could make harmful interventions appear protective. 55–57 This is an issue that was difficult to overcome with the data that were available, as reliably imputing the assigned intervention at baseline is difficult given that the excluded population is likely to be a quite distinct population from the included population. Nevertheless, selection bias is unlikely to be solely responsible for the observed effect, as the Kaplan–Meier curves for ischaemic outcomes (see Figure 10) diverge for the entire follow-up period; the inclusion of the population with early events would have influenced the curves only in the first few months.

It is also important to note that there is considerable uncertainty about the benefits of adding a P2Y₁₂ inhibitor to aspirin monotherapy after CABG. There are no large, pragmatic, multicentre RCTs of DAPT versus aspirin monotherapy in CABG populations, and most of the RCTs included in the meta-analyses summarised above52–54 were at high risk of bias and had saphenous vein graft failure as a primary outcome. Saphenous vein graft failure is not a clinical end point and does not correlate well with hard clinical end points (mortality, MI and repeat intervention). 58

Our analysis was intention to treat, so patients were analysed according to intervention groups assigned at baseline, regardless of adherence or switches in antiplatelet treatment. Generally, studies report that between 45% and 65% of patients adhere to medications prescribed for secondary prevention, with few differences between drug classes. 59–61 Several studies have highlighted potential adherence issues among patients taking antiplatelet therapy/anticoagulants59–61 and studies conducted in real-world settings suggest that non-adherence to DAPT is a common problem, affecting up to 48% of patients. 59 This study reflects this, showing that non-adherence in the DAPT group was 46%, which was much higher than in the aspirin group (30%). Furthermore, over one-quarter of all patients in the DAPT group were identified as having switched from DAPT, meaning that they stopped aspirin, clopidogrel or both. Just under half of all MIs in the AC group (n = 6/13) occurred among the switchers. It is possible that, in this study, non-adherence in a high-risk population prescribed DAPT contributed to the high incidence of MACEs that we observed.

Trends in antiplatelet prescribing and rates of bleeding over time

Figure 11 describes how the target trial population was assembled from the available data sets. Figure 12 shows the trends in antiplatelet prescriptions and rates of bleeding between 2010 and 2017 in the target trial population. There was a slight decrease in both aspirin and DAPT (AC) prescriptions over time (from 30.2% in 2010 to 23.5% in 2017 for aspirin and from 42.9% in 2010 to 32.7% in 2017 for AC). There was also a slight increase in the number of patients being prescribed no antiplatelet therapy or being prescribed some other regimen (e.g. one or more P2Y₁₂ inhibitors). Both major and minor bleeding rates decreased, from 50.0 and 80.0 events, respectively, per 1000 person-years in 2010 to 13.7 and 51.3 events, respectively, per 1000 person-years in 2017.

FIGURE 11.

Flow diagram describing the construction of the conservatively managed ACS target trial. C, clopidogrel; P, prasugrel; SA, sensitivity analysis; T, ticagrelor.

FIGURE 12.

Proportion of conservatively managed ACS patients being prescribed different antiplatelet regimes by year and rates of major (HES), minor (CPRD) and total bleeding events by year among conservatively managed ACS patients. (a) Different antiplatelet regimes; and (b) bleeding events.

Baseline characteristics of participants included in and those excluded from the target trial

Table 18 shows the baseline characteristics of patients included in and those excluded from the primary analysis. Of the 15,989 patients with linked CPRD–HES data, 10,943 (68%) were eligible and included in the target trial; 4357 of these patients (40%) were excluded because they could not be assigned to an intervention group. The number of eligible patients decreased every year between 2010 and 2017 (as a result of the decline over time in the number of practices in the CPRD GOLD). 50 The covariates with an imbalance between the aspirin and the AC groups were smoking (14% were smokers in the aspirin group and 18% were smokers in the AC group, SMD 0.11), history of MI (39% in the aspirin group vs. 57% in the AC group, SMD 0.38), history of CABG/PCI (21% in the aspirin group vs. 14% in the AC group, SMD 0.20), hypercholesterolaemia (17% in the aspirin group vs. 11% in the AC group, SMD 0.17) and use of PPIs (51% in the aspirin group vs. 43% in the AC group, SMD 0.15). For patients with unknown intervention (excluded from the primary analysis), compared with patients with known intervention (included in the primary analysis), the covariates with an imbalance were history of MI (50% in the included population and 35% in the excluded population, SMD 0.32), history of CABG/PCI (17% in the included population vs. 8% in the excluded population, SMD 0.26) and history of IHD (77% in the included population vs. 62% in the excluded population, SMD 0.31). There was a difference of 2 days in length of hospital stay between the aspirin and the AC groups [median 3 (IQR 1–7) days and median 5 (IQR 3–9) days, respectively].

Of the 4357 patients who could not be assigned to an intervention group, 2293 (53%) either died or had a major bleed or ACS event before their first prescription, and 2064 (47%) had no prescription for an antiplatelet agent in the 2 months after their index event. The characteristics of patients who died or had a major bleed or ACS event compared with those who had no antiplatelet prescription within 2 months of discharge are shown in Table 19. There were large differences in age (79 vs. 67 years, respectively, SMD 0.83) and history of MI (42% vs. 26%, respectively, SMD 0.34). There were also differences in BMI (26.6 vs. 28.1 kg/m², respectively, SMD 0.25) and in the proportion of patients with heart failure (15% vs. 8%, respectively, SMD 0.22), renal disease (13% vs. 7%, respectively, SMD 0.20), cancer (12% vs. 6%, respectively, SMD 0.18) and diabetes (26% vs. 20%, respectively, SMD 0.16). There were also differences in ethnic group (6% other than white vs. 9% other than white, respectively, SMD 0.13), the proportion of current smokers (16% vs. 18%, respectively, SMD 0.13) and the proportion of patients with hypercholesterolaemia (9% vs. 13%, respectively, SMD 0.13) and anaemia (10% vs. 7%, respectively, SMD 0.11).

Bleeding events among participants included in and those excluded from the target trial

Of the 6586 patients included in the target trial, 688 (10%) experienced at least one bleeding event: 216 out of 2609 (8%) in the aspirin group and 472 out of 3977 (12%) in the AC group. With regard to major and minor bleeding events, 290 out of 6586 (4%) patients experienced a major bleed and 463 out of 6586 (7%) experienced a minor bleed. The proportions of patients experiencing a major and a minor bleeding event in the aspirin group were 96 out of 2609 (4%) and 141 out of 2609 (5%), respectively, whereas, in the AC group, the proportions of patients experiencing a major and a minor bleeding event were 194 out of 3977 (5%) and 322 out of 3977 (8%), respectively.

Figure 13 shows the Kaplan–Meier curves of cumulative bleeding (any bleed, major bleed and minor bleed) in the AC compared with the aspirin groups. The cumulative incidence of any bleeding increased steadily over the 12 months, but was higher in the AC group than in the aspirin group. The survival curves crossed after approximately 35 days, with a lower incidence of bleeding in the AC group before this point. This was reflected in the survival curves for both major and minor bleeding.

FIGURE 13.

Kaplan–Meier curves displaying cumulative bleeding according to intervention group. (a) Any bleeding; (b) major bleeding; and (c) minor bleeding. Plots are weighted according to the inverse probability of treatment received, and so compare outcomes if all eligible patients received aspirin or AC.

The crude incidence rates of major and minor bleeds were 24% higher (38 vs. 50 events per 1000 person-years) and 49% higher (56 vs. 84 events per 1000 person-years), respectively, in the AC group than in the aspirin group (Table 20). Of those who experienced a bleeding event within 12 months, the majority (489/688; 71%) of patients experienced only one bleeding event; 132 out of 688 (19%) experienced two bleeding events, and the remainder (67/688; 10%) experienced three or more bleeds. Over 40% of bleeds were gastrointestinal in origin; skin or soft-tissue bleeds and ear, nose and throat bleeds each accounted for just under one-fifth of bleeds (Table 21). More participants in the aspirin group than in the AC group had gastrointestinal bleeds (54% vs. 39%, respectively), but slightly fewer had skin or soft-tissue bleeds (15% vs. 21%, respectively) and ear, nose and throat bleeds (10% vs. 22%, respectively).

Patients who could not be assigned an intervention because they experienced a bleed or ischaemic event or died before their first prescription, or because they had no prescription in the CPRD within 2 months of discharge (40% of the eligible population), had a lower bleeding rate than the patients included in the target trial (7% vs. 10%, respectively) (Table 22). The bleeding rate was slightly higher among those who had no prescription in the CPRD within 2 months of discharge than among those who experienced an event or died (9% vs. 6%, respectively).

Analyses for the primary outcome (bleeding)

The primary analysis excluded patients to whom we could not assign an intervention (n = 4357, 40% of the eligible population). The crude and adjusted HRs indicated an increase of about 40% in the hazard of bleeding in the AC group compared with the aspirin group (HR 1.44, 95% CI 1.23 to 1.70, and HR 1.43, 95% CI 1.21 to 1.69, respectively) (see Table 22). When split into major and minor bleeding, the hazard of major bleeding increased by 34% (HR 1.34, 95% CI 1.04 to 1.72), and the hazard of minor bleeding increased by 51% (HR 1.51, 95% CI 1.23 to 1.86) in the AC group compared with the aspirin group.

Mortality and ischaemic events among participants included in and those excluded from the target trial

Figure 14 shows the Kaplan–Meier curves for the secondary outcomes of all-cause and cardiovascular mortality, mortality from bleeding, MI, stroke, additional coronary intervention and the composite outcome of MACE.

FIGURE 14.

Kaplan–Meier curves displaying cumulative all-cause and cardiovascular mortality, mortality from bleeding and ischaemic events, according to intervention group. (a) All-cause mortality; (b) cardiovascular mortality; (c) mortality from bleeding; (d) MI; (e) stroke; (f) additional PCI; and (g) MACE.

There were large differences in mortality and ischaemic events between patients included in the target trial and patients who were eligible for inclusion but were not included (40% of all eligible patients) (Table 23), although these were driven entirely by the group of patients excluded because they had a bleeding or ischaemic event prior to their first prescription in the CPRD.

Analyses for the secondary outcomes (mortality and ischaemic events)

There was no association between antiplatelet prescription (AC vs. aspirin) and all-cause mortality, cardiovascular mortality, mortality from bleeding or stroke (see Table 23).

Treatment switches and adherence

Treatment switches are shown in Table 24. In the 12 months after the index event, 608 out of 2609 (23%) patients in the aspirin group were identified as ‘switchers’. There were 657 treatment switches; 431 (66%) initiated a second antiplatelet and 226 (34%) stopped aspirin. The median time to switching was 5 months among those who initiated a second antiplatelet and 8 months among those who stopped aspirin. On average, patients who initiated a second antiplatelet received 6.5 prescriptions of the second antiplatelet (6 months’ supply).

Among patients assigned AC, 668 out of 3977 (24%) were identified as switchers in the 12 months after the index event. There were 986 treatment switches; of these, 531 (54%) were aspirin discontinuations; 269 (27%) were clopidogrel discontinuations, 156 (16%) were aspirin and clopidogrel discontinuations and 30 (3%) were initiations of a different P2Y₁₂ inhibitor. The median time to switching was between 5 and 8 months in all groups of switchers.

Across the groups, 283 switchers had a bleed or ischaemic event, 108 (38%) in aspirin and 175 (62%) in AC. Most of these events occurred before the switch. In the aspirin group, 12% of those who discontinued aspirin had a bleeding or ischaemic event, but 35% who initiated a second antiplatelet had a bleeding (33 patients) or ischaemic event (55 patients). The numbers of bleeding events before and after switching were similar in both groups of switchers, although the number of ischaemic events was highest before the switch among those who initiated a second antiplatelet, suggesting that the ischaemic event triggered the switch.

In the AC group, the proportion of patients experiencing bleeding and ischaemic events was highest among the switchers who initiated a different P2Y₁₂ inhibitor (60%, 16 ischaemic events, all before the switch, and five bleeding events), but was also relatively high among those who discontinued both aspirin and clopidogrel (32%, 27 bleeds and 29 ischaemic events); it was lowest among the switchers who discontinued aspirin only (26%, 78 bleeds and 70 ischaemic events). Across all groups of switchers, a higher proportion experienced both bleeding and ischaemic events before, rather than after, the switch.

Adherence, defined as a MPR of ≥ 0.8, was 56% in the aspirin group and 60% in the AC group.

Discussion

In the conservatively managed ACS target trial, including 6586 patients, the overall rate of bleeding was 10%, whereas the rates of major and minor bleeding were 4% and 7%, respectively. The rates of major bleeding with aspirin and DAPT with clopidogrel that we observed in our population (4% for aspirin and 5% with DAPT) were slightly higher than, but comparable to, those reported in the conservatively managed ACS population (n = 7985) in the Clopidogrel in Unstable Angina to Prevent Recurrent Events (CURE) RCT: 3% for aspirin and 4% for DAPT with AC. 62,63

The main finding in our conservatively managed ACS target trial was that, compared with aspirin, DAPT with clopidogrel increased the risk of any bleed by 43%, and the risk of major and minor bleeding by 34% and 51%, respectively. None of the sensitivity analyses markedly attenuated these HRs for bleeding. There was no evidence of subgroup effects, although a concurrent prescription for PPIs attenuated the effect of DAPT on any bleeding events. In the CURE trial, DAPT with clopidogrel also increased the risk of major bleeding by 38% (RR 1.38, 95% CI 1.13 to 1.67), which was comparable to the increase in risk we observed in our population.

Similar to the finding in the CABG target trial, DAPT with clopidogrel increased the hazards of MI, additional coronary intervention and MACEs by 88%, 86% and 58%, respectively, compared with aspirin. This is in contrast to the finding in the CURE RCT, which showed that DAPT with clopidogrel was associated with a 20% reduction in the combined end point of cardiovascular mortality, non-fatal MI and stroke (9% vs. 11%, RR 0.80, 95% CI 0.69 to 0.92). Other RCTs in ACS populations show a 20% decrease in the risk of secondary ischaemic events with DAPT with clopidogrel, regardless of revascularisation status. 64,65

Potential reasons for the discrepancy between our findings and those of the CURE trial are confounding, selection bias and treatment switches/adherence. Most baseline characteristics were reasonably balanced between the aspirin and the DAPT groups, and the risk factors for ischaemia that showed an imbalance between the aspirin and DAPT groups (smoking, hypercholesterolaemia, previous MI and revascularisation) were not uniformly in one direction, that is not always higher in the AC group (the proportion of smokers and of those with a history of MI was greater in the DAPT group, whereas the proportion of those who had a previous revascularisation and hypercholesterolaemia was higher in the aspirin group). These factors were all adjusted for in the analysis. Furthermore, unmeasured confounding is likely to be less of an issue in the conservatively managed ACS target trial than in the CABG or PCI target trials because there are no definitive clinical guidelines to guide antiplatelet selection and no procedure-specific characteristics to influence DAPT prescribing. Nevertheless, we cannot rule out that patients perceived to be at high risk of secondary ischaemic events were more likely to be prescribed DAPT, and indeed those in the AC group had a longer median length of stay (by 2 days) than those in the aspirin group, suggesting that the former had more significant ACS events.

Selection bias may also be an explanation for the observed results, as selection for the target trial is likely to be associated with both the assignment to intervention and the outcome. We excluded 40% (4357/10,943) of the eligible population because they experienced a major bleed or ischaemic event before their first prescription or because they had no prescription within 2 months of hospital discharge. The former (just over half of the excluded population, 2293/4357) were much older, with a higher incidence of previous MI, heart failure, renal disease and cancer, and so were likely to be at higher risk of secondary ischaemic events. An older, more frail population is more likely to be prescribed aspirin. It is possible that a true higher risk of MACE in the aspirin group may be masked by excluding this susceptible population from the risk set. Although we imputed intervention status at baseline in a sensitivity analysis, this was driven by the characteristics of patients included in the target trial, and, therefore, may not truly reflect the actual prescription at baseline. It is debatable the extent to which imputation for such a large number of missing data is effective. The extent to which selection bias is responsible for the observed results is not clear. It is worth noting that the survival curves for MACE (see Figure 14) continue to diverge until the end of follow-up, suggesting a true higher risk of MACE in the included population, rather than an effect driven by the exclusion of eligible participants with an early event, which would have affected the shape of the survival curves early during follow-up, but not later on.

Non-adherence was 40% in the DAPT group and 44% in the aspirin-only group. These DAPT adherence rates reflect those reported by other studies in real-world populations59–61 and mirror those in the CABG target trial. Over one-quarter of patients in the conservatively managed ACS target trial switched prescription (either stopped aspirin or initiated a second antiplatelet agent in the aspirin group, or stopped aspirin or clopidogrel or both aspirin and clopidogrel in the DAPT group). The proportion of patients experiencing bleeding and ischaemic events was highest among the switchers who discontinued both aspirin and clopidogrel (32%). It is possible that non-adherence in the higher-risk population prescribed DAPT contributed to the high incidence of MACEs that we observed.

Trends in antiplatelet prescribing and rates of bleeding over time

Figure 15 shows how the target trial population was assembled. Figure 16 show the trends in antiplatelet prescriptions and rates of bleeding between 2010 and 2017. There was a large decrease in AC (from 80.4% in 2010 to 30.9% in 2017) and a large increase in AT (from 0% and 0.4% in 2010/11, to 12.4% in 2012 and 54.4% in 2017) over time. There was a small increase in AP between 2010 and 2013, followed by a consistent decrease thereafter. Other prescriptions (P2Y₁₂ inhibitor monotherapy, aspirin only and other, e.g. more than one P2Y₁₂ inhibitor) remained steady, but were below 7%. Major bleeding rates were similar over time, at 29.6 and 26.7 events per 1000 person-years in 2010 and 2017, respectively. Minor bleeding rates increased slightly, from 75.1 to 89.5 events per 1000 person-years, in 2010 and 2017, respectively.

FIGURE 15.

Flow diagram describing the construction of the emergency PCI target trial. C, clopidogrel; P, prasugrel; SA, sensitivity analysis; T, ticagrelor.

FIGURE 16.

Proportion of emergency PCI patients being prescribed different antiplatelet regimes by year and rates of major (HES), minor (CPRD) and total bleeding events by year among emergency PCI patients. (a) Different antiplatelet regimes; and (b) bleeding events.

Baseline characteristics of participants included in and those excluded from the target trial

Table 25 shows the baseline characteristics of participants included in and those excluded from the target trial. Of the 11,361 patients with linked CPRD–HES data, 5738 (51%) were eligible and included, and 520 (9%) were excluded because they could not be assigned to an intervention group. The covariates with an imbalance between the AC and the AT groups were age (mean age of 66.1 years in the AC group vs. 62.5 years in the AT group, SMD 0.3), smoking (27% were smokers in the AC group vs. 34% in the AT group, SMD 0.15), history of IHD (90% in the AC group vs. 83% in the AT group, SMD 0.19), hypertension (43% in the AC group vs. 35% in the AT group, SMD 0.17), hypercholesterolaemia (22% in the AC group vs. 16% in the AT group, SMD 0.16), peripheral vascular disease (5% in the AC group vs. 3% in the AT group, SMD 0.11) and renal disease (5% in the AC group vs. 3% in the AT group, SMD 0.11). There was no difference in median length of hospital stay between the AC and the AT groups [median 2 (IQR 1–3) days for both groups].

For patients excluded from the primary analysis, compared with patients included, the covariates with an imbalance were age (mean 64.0 years in the included population vs. 67.5 years in the excluded population, SMD 0.26), BMI (mean 28.4 kg/m² in the included population vs. 27.4 kg/m² in the excluded population, SMD 0.18), history of MI (77% in the included population vs. 61% in the excluded population, SMD 0.35), history of IHD (86% in the included population vs. 68% in the excluded population, SMD 0.43), hypercholesterolaemia (18% in the included population vs. 13% in the excluded population, SMD 0.14), heart failure (7% in the included population vs. 11% in the excluded population, SMD 0.16), cancer (4% in the included population vs. 8% in the excluded population, SMD 0.15) and prescription for non-steroidal anti-inflammatory drugs (NSAIDs) (19% in the included population vs. 14% in the excluded population, SMD 0.13).

Of the 520 patients without intervention, 250 (48%) died or had a major bleed or ACS event before their first prescription, and 270 (52%) had no prescription for an antiplatelet agent in the 2 months after their index event. The characteristics of patients who died or had a major bleed or ACS event, compared with those who had no antiplatelet prescription within 2 months of discharge, are shown in Table 26. The former were older (72 years vs. 63 years, respectively, SMD 0.70); had a greater proportion of women (33% vs. 23%, respectively, SMD 0.22); had fewer smokers (28% vs. 39%, respectively, SMD 0.25); had greater proportions of patients with a history of CABG/PCI (37% vs. 26%, respectively, SMD 0.25), previous surgery (7% vs. 3%, respectively, SMD 0.16), diabetes (24% vs. 12%, respectively, SMD 0.30), hypertension (43% vs. 29%, respectively, SMD 0.30), peripheral vascular disease (6% vs. 4%, respectively, SMD 0.11), heart failure (14% vs. 9%, respectively, SMD 0.15), renal disease (99% vs. 3%, respectively, SMD 0.26) and liver cirrhosis (1% vs. 0%, respectively, SMD 0.13); and a smaller proportion of patients with previous bleeding (7% vs. 3%, respectively, SMD 0.16). Patients who died or had an event also had more prescriptions of steroids, PPIs and anticoagulants (all SMDs > 0.1).

Bleeding events among participants included in and those excluded from the target trial

In the emergency PCI target trial, comprising all patients with ACS undergoing emergency PCI (3845 STEMI, 3082 NSTEMI and 4186 unstable angina patients), we compared AC with AT. AP was prescribed to ACS STEMI patients only; therefore, the comparison of AC with AP was restricted to the STEMI population. Of the 4689 patients prescribed AC or AT, 416 (9%) experienced at least one bleeding event: 209 out of 2769 (8%) patients prescribed AC and 207 out of 1920 (11%) patients prescribed AT. With regard to major and minor bleeding events, 117 out of 4689 (3%) patients experienced a major bleed and 332 out of 4689 (7%) experienced a minor bleed. The proportion of patients experiencing a major or a minor bleeding event was 63 out of 2769 (2%) and 161 out of 2769 (6%), respectively, in the AC group and 54 out of 1920 (3%) and 171 out of 1920 (9%), respectively, in the AT group.

Figure 17 shows the Kaplan–Meier curves of cumulative bleeding events [any, major (HES) and minor (CPRD)] in the AC and AT groups. The cumulative incidence of any bleeding increased steadily over the 12 months, but was higher in the AT group than in the AC group. The survival curves crossed twice. Major bleeds were initially more frequent in the AT group, until approximately 50 days, then were more frequent in the AC group, until approximately 200 days (6.5 months), and thereafter were more frequent in the AT group. Minor bleeds were consistently more frequent in the AT group than in the AC group.

FIGURE 17.

Kaplan–Meier curves displaying cumulative bleeding according to intervention group. (a) Any bleeding; (b) major bleeding; and (c) minor bleeding. Plots are weighted according to the inverse probability of treatment received, and so compare outcomes if all eligible patients received AC or AT.

In the AT group, compared with the AC group, the crude incidence rates of major and of minor bleeds were 24% higher (29 vs. 23 events per 1000 person-years) and 56% higher (94 vs. 60 events per 1000 person-years), respectively (Table 27). Of those who experienced a bleeding event within 12 months, the majority of patients experienced only one bleeding event (310/416, 75%), 78 out of 416 (19%) experienced two bleeding events, and the remainder (28/416, 7%) experienced three or more bleeds. Bleeds by site are shown in Table 28; there were no major differences between intervention groups.

Patients who could not be assigned an intervention because they experienced a bleed or ischaemic event or died before their first prescription, or because they had no prescription in the CPRD within 2 months of discharge (9% of the eligible population), had a lower bleeding rate than the patients included in the target trial (3% vs. 9%) (see Table 29). The bleeding rate was slightly higher among those who had no prescription in the CPRD within 2 months of discharge than among those who experienced an event or died (3% vs. 2%).

Analyses for the primary outcome (bleeding)

The primary analysis excluded patients for whom we could not assign an intervention (n = 4689). The crude and adjusted HRs indicated an increase of about 50% in the hazard of bleeding in the AT group compared with the AC group (HR 1.48, 95% CI 1.22 to 1.80, and HR 1.47, 95% CI 1.19 to 1.82, respectively) (Table 29). When split into major and minor bleeding, there was a 33% increased hazard of major bleeding (HR 1.33, 95% CI 0.89 to 1.99) and a 60% increased hazard of minor bleeding (HR 1.60, 95% CI 1.26 to 2.03) in the AT group compared with the AC group.

Mortality and ischaemic events among participants included in and those excluded from the target trial

Figure 18 shows the Kaplan–Meier curves for the secondary outcomes of all-cause and cardiovascular mortality, mortality from bleeding, MI, stroke, additional coronary intervention and the composite outcome of MACE. There were large differences in mortality and ischaemic events between patients included in the target trial and patients who were eligible for inclusion but were not included (9% of all eligible patients) (see Table 30).

FIGURE 18.

Kaplan–Meier curves displaying cumulative all-cause and cardiovascular mortality, mortality from bleeding and ischaemic events, according to intervention group. (a) All-cause mortality; (b) cardiovascular mortality; (c) mortality from bleeding; (d) MI; (e) stroke; (f) additional PCI; and (g) MACE.

Analyses for the secondary outcomes (mortality and ischaemic events)

There was no association between antiplatelet prescription (AT vs. AC) and any of the secondary outcomes (see Table 30).

Treatment switches and adherence

Table 31 shows treatment switches in the emergency PCI population by intervention group (AC and AT), and by type of switch and whether the switch occurred before or after a bleeding or ischaemic event.

In the 12 months after the index event, 379 out of 2769 (14%) patients in the AC group were identified as ‘switchers’. There were 560 treatment switches; 300 of these (54%) were aspirin discontinuations, 124 (22%) were clopidogrel discontinuations, 84 (15%) were aspirin and clopidogrel discontinuations and 52 (9%) were initiations of a different P2Y₁₂ inhibitor.

Among patients assigned AT, 404 out of 1920 (21%) were identified as switchers. There were 454 treatment switches; of these, 200 (44%) were aspirin discontinuations, 154 (34%) were ticagrelor discontinuations, 85 (19%) were aspirin and ticagrelor discontinuations and 15 (3%) were initiations of a different P2Y₁₂ inhibitor.

The median time to switch was around 8 months in all groups of switchers. Across AC and AT, 125 switchers had a bleed or ischaemic events, 65 (52%) in AC and 60 (48%) in AT. Most of these events occurred before the switch.

In all intervention groups, the number of ischaemic events was larger among those who switched, compared with event rates in the population overall. Adherence, defined as a MPR of ≥ 0.8, was 71% in the AC group, 69% in the AP group and 68% in the AT group.

Trends in antiplatelet prescribing and rates of bleeding over time

The target trial population is shown in Figure 15. Trends in antiplatelet prescriptions and rates of bleeding between 2010 and 2017 are shown in Figure 19. There was a large decrease in DAPT prescriptions with AC (from 74.5% in 2010 to 16.7% in 2017) and a large increase in DAPT prescriptions with AT (from 0% and 0.6% in 2010/11, to 14.6% in 2012 and 67.0% in 2017) over time. Prescriptions of DAPT with AP increased from 13.3% in 2010 to 21.3% in 2011, remained at this level until 2012 and then decreased thereafter to reach 5.9% in 2017. Prescriptions of aspirin and P2Y₁₂ inhibitor monotherapy remained steady, and a small proportion of patients (about 5% over 2010–17) received no antiplatelet prescription at all. Despite the large increase in AT prescriptions over time, major bleeding rates increased only marginally, from 28.4 events per 1000 person-years to 36.4 events per 1000 person-years in 2017. Minor bleeding rates increased over time, from 73.8 to 113.8 events per 1000 person-years in 2010 and 2017, respectively.

FIGURE 19.

Proportion of emergency PCI STEMI patients being prescribed different antiplatelet regimes by year and rates of major (HES), minor (CPRD) and total bleeding events by year among emergency PCI STEMI patients. (a) Different antiplatelet regimes; and (b) bleeding events.

Baseline characteristics of participants included in and those excluded from the target trial

Table 32 shows the baseline characteristics of patients included in and those excluded from the primary analysis. Of the 5738 patients with linked CPRD–HES data and eligible for the emergency PCI analysis, 2893 (50%) were STEMI patients. Of these patients, 306 (11%) were excluded because they could not be assigned to an intervention group.

Compared with patients in the AC group, the population in the AP group was younger; had a higher proportion of men; had more smokers; was less likely to have had a history of MI and IHD; and was also less likely to have diabetes, hypertension, hypercholesterolaemia and renal disease (all SMDs > 0.10). Patients in the AT group were also younger, with more smokers than in the AC group, but were otherwise balanced with regard to all other baseline characteristics. There was no difference in median length of stay between the AT, the AP and the AC groups [3 (IQR 2–4) days, 3 (IQR 2–3) days and 3 (IQR 2–3) days, respectively].

Of the 306 patients without intervention, 161 (53%) either died or had a major bleed or ACS event before their first prescription, and 145 (47%) had no prescription for an antiplatelet agent in the 2 months after their index event. Compared with the 2587 out of 2893 (89%) patients with known intervention (included in the primary analysis), the population with unknown intervention (excluded from the primary analysis) was older; had a higher proportion of women; had more smokers/ex-smokers; and had a higher proportion of patients with a history of CABG/PCI, previous surgery, a history of IHD, hypertension, heart failure, renal disease, cancer and liver cirrhosis (see Table 32). Fewer patients with unknown intervention were taking NSAIDs, but more of them were taking anticoagulants.

Compared with the population who had no prescription in the CPRD within 2 months of discharge, the population who had a bleed or ischaemic event was older; had a higher proportion of women; had fewer smokers; had a higher proportion of patients with a history of CABG/PCI, previous surgery, history of IHD, diabetes, hypertension, heart failure, renal disease and cancer; and had a higher proportion of patients using NSAIDs, steroids, PPIs and anticoagulants (Table 33).

Bleeding events among participants included in and those excluded from the target trial

Of the 2587 STEMI patients, 260 (10%) experienced at least one bleeding event: 80 out of 1023 (8%) in the AC group, 46 out of 406 (11%) in the AP group and 134 out of 1158 (12%) in the AT group. With regard to major and minor bleeding events, 70 out of 2587 (3%) patients experienced a major bleed and 208 out of 2587 (8%) experienced a minor bleed. The proportions of patients experiencing a major and a minor bleeding event were 22 out of 1023 (2%) and 62 out of 1023 (6%), respectively, in the AC group; 9 out of 406 (2%) and 39 out of 406 (10%), respectively, in the AP group; and 39 out of 1158 (3%) and 107 out of 1158 (9%), respectively, in the AT group.

Figure 20 shows the Kaplan–Meier curves of cumulative bleeding (any bleed) in the AC versus the AP versus the AT groups. The cumulative incidence of any bleeding increased steadily over the 12 months, but was higher in the AP and AT groups than in the AC group. The number of major bleeds was larger in the AT group and the numbers of minor bleeds were larger in the AP and AT groups than in the AC group. The crude incidence rates of major and minor bleeds were 3% higher (23 vs. 22 events per 1000 person-years) and 62% higher (102 vs. 63 events per 1000 person-years) in the AP group than in the AC group, and were 58% higher (34 vs. 22 events per 1000 person-years) and 54% higher (97 vs. 63 events per 1000 person-years) in the AT group than in the AC group (Table 34). Of those who experienced a bleeding event within 12 months, the majority of patients experienced only one bleeding event (187/260, 72%); 57 out of 260 (22%) experienced two bleeding events; and the remainder (16/260, 6%) experienced three or more bleeds. Bleeds by site are shown in Table 35; there were slightly larger numbers of ear, nose and throat bleeds in the AP group than in the AC or AT groups, and a larger number of gastrointestinal bleeds in the AC group than in the AP or AT groups.

FIGURE 20.

Kaplan–Meier curves displaying cumulative bleeding according to intervention group. (a) Any bleeding; (b) major bleeding; and (c) minor bleeding. Plots are weighted according to the inverse probability of treatment received, and so compare outcomes if all eligible patients received AC, AP or AT.

Patients who could not be assigned an intervention because they experienced a bleed or ischaemic event or died before their first prescription or because they had no prescription in the CPRD within 2 months of discharge (11% of the eligible population) had a lower bleeding rate than the patients included in the target trial (3% vs. 10%, respectively) (Table 36). The bleeding rate was slightly higher among those who had no prescription in the CPRD within 2 months of discharge than among those who experienced an event or died (4% vs. 2%, respectively).

Analyses for the primary outcome (bleeding)

The primary analysis excluded patients for whom we could not assign an intervention (306/2993, 11%). The crude and adjusted HRs indicated an increase in the hazard of bleeding in the AP group compared with the AC group (HR 1.48, 95% CI 1.02 to 2.12, and HR 1.77, 95% CI 1.21 to 2.59, respectively). The crude and adjusted HRs also indicated an increase in the hazard of bleeding in the AT group compared with the AC group (HR 1.53, 95% CI 1.16 to 2.01) and HR 1.50, 95% CI 1.10 to 2.05, respectively) (see Table 36). When split by major and minor bleeding, there was an increased hazard of major bleeding for both the AP versus the AC groups (HR 1.33, 95% CI 0.59 to 3.01) and the AT versus the AC groups (HR 1.86, 95% CI 1.05 to 3.32), although the estimate for AP versus AC is quite imprecise. The hazards for minor bleeding increased for the AP group, compared with the AC group (HR 1.86, 95% CI 1.23 to 2.82), and for the AT group, compared with the AC group (HR 1.49, 95% CI 1.06 to 2.09).

Mortality and ischaemic events among participants included in and those excluded from the target trial

Figure 21 shows the Kaplan–Meier curves for the secondary outcomes of all-cause and cardiovascular mortality, mortality from bleeding, MI, stroke, additional coronary intervention and the composite outcome of MACE. Patients not included in the target trial (11% of all eligible patients) had higher event rates than those included, although these were driven largely by the group of patients excluded because they had a bleeding or ischaemic event prior to first prescription in the CPRD (Table 37). The 145 patients excluded because they had no prescription in the CPRD within 2 months of discharge had event rates comparable to those of the included population.

FIGURE 21.

Kaplan–Meier curves displaying cumulative all-cause and cardiovascular mortality, mortality from bleeding and ischaemic events, according to intervention group. (a) All-cause mortality; (b) cardiovascular mortality; (c) mortality from bleeding; (d) MI; (e) stroke; (f) additional PCI; and (g) MACE.

Analyses for the secondary outcomes (mortality and ischaemic events)

There was no association between antiplatelet prescription (AP vs. AC and AT vs. AC) and any outcome (see Table 37).

Treatment switches and adherence

In the AC group, 141 out of 1023 (14%) patients were identified as ‘switchers’. There were 205 treatment switches (Table 38). Of these, 114 (56%) were aspirin discontinuations, 43 (21%) were clopidogrel discontinuations, 30 (15%) were aspirin and clopidogrel discontinuations and 18 (9%) were initiations of a different P2Y₁₂ inhibitor. The median time to switch was between 7 and 8 months, although those who initiated a different P2Y₁₂ inhibitor switched at a median time of 1 month.