This work was designed as a prospective follow-up study. All patients diagnosed with ACS upon discharge and at the age of 18 years or older were enrolled at Tongji Hospital, Tongji Medical College in Wuhan from November 5, 2013 to December 31, 2014. ACS involves MI (NSTEMI and STEMI) and UA. MI is defined as cardiomyocyte necrosis in clinical setting and is consistent with acute myocardial ischemia; UA is defined as myocardial ischemia at rest or minimal exertion without cardiomyocyte necrosis [2]. Excluded from the study were patients with metastatic cancers or died prior to 30 days after discharge or who were not followed up.

We developed an initial case report form and subsequently finalized it for field use after pilot testing. Information was collected using baseline characteristics on demographics (age, gender, education level, and types of medical insurance), comorbid conditions, history of atrial fibrillation, heart failure, stroke, implantable cardioverter defibrillator (ICD), radiofrequency ablation, chronic renal dysfunction (CRD), types of ACS (STEMI, non-STEMI, and UA), and ACS treatment, which consisted of PCI, coronary artery bypass grafting (CABG), combined PCI and CABG treatment (hybrid procedure), and conservative treatment from medical records or personal interviews with patients. Case report forms were completed by trained hospital staff using EpiData Software Version 3.1. A unique and permanent civil number was used to identify the patients enrolled.

All patients who enrolled voluntarily signed written informed consents and were interviewed by phone and face-to-face by trained researchers at the 1st, 6th, and 12th month after discharge using a standardized questionnaire to determine vital status, current medications (patients were asked to collect all their current medications and to read out every brand name), date, and reason for hospital readmissions. When a patient died, a close relative was interviewed. In case of readmission, clinical records were reviewed to establish reasons for readmission and medications during hospitalization and at discharge.

We registered each patient’s discharge medications, including aspirin, P2Y12 antagonists, statins, ACEIs/ARBs, and b-blockers from medical records. P2Y12 antagonists in this study were clopidogrel hydrogen sulfate, ticagrelor, and clopidogrel bisulfate. The statins were atorvastatin calcium (produced by Pfizer or Jarlin in Beijing), rosuvastatin calcium, and others. The ACEIs/ARBs were ACERTIL (perindopril), fosinopril sodium, benazepril hydrochloride, perindopril and indapamide (BIPREL forte), valsartan capsules, telmisartan, irbesartan, and others. The b-blockers were metoprolol succinate sustained-release tablets and Concor (bisoprolol fumarate). We assessed the adherence to each medication of each patient by calculating the proportion of days covered (PDC), which was considered to measure patients’ ATM [8]. The PDC for each medication is defined as the days that patients were provided with filled prescriptions divided by the 12-month observational time interval or first day after discharge to the day of the last interview. The days when patients were given filled prescriptions started from the first day after discharge to the last day, unless they stopped taking medications for at least three months. From phone and face-to-face interviews and from reviews of clinical records, we recorded discontinued medications and approximate date of interruption. Thus, the approximate PDC could be determined for each patient and each medication. We also classified the patients under low adherence (PDC<50%) and high adherence (PDC>50%) groups.

We recorded the occurrence of MACE for 12 months after discharge given that one year is the recommended duration for effects of P2Y12 antagonists. MACE was defined as the first occurrence of cardiovascular death, MI, and revascularization.

For continuous variables, results were expressed as mean values±SD. We used frequency statistics to summarize baseline population characteristics (demographics, medical insurance types, comorbid conditions, ACS types, ACS treatment types, cardiovascular history, and concomitant treatment). Chi-square test was used to compare groups and to identify significant differences in their adherence to each medication. Regression analysis was conducted using PDC as bivariate measurement (low adherence vs. high adherence). Multivariate logistic regression was used to evaluate independent associations between concomitant medications and adherence to aspirin, P2Y12 antagonists, ACEIs/ARBs and b-blockers after adjustment for ACS treatments for PCI, CABG, combined PCI and CABG, and conservative treatment, respectively, using forward: LR. For statins, adherence was adjusted for ACS types and radiofrequency ablation.

We used Kaplan–Meier curves to determine one-year cardiac MACE rates and Cox regression modeling to assess whether low ATM was associated with higher risk of MACE. Hazard ratios were adjusted for potential confounders (ACS treatment, radiofrequency ablation, and concomitant treatments) using multivariate Cox regression analysis (forward: LR). For the PCI population, hazard ratios were merely adjusted by concomitant treatments. Finally, data were assessed by power analysis (a = 0.05) using PASS Version 11.

All statistical calculations were performed using the Statistical Package for Social Sciences (SPSS) Version 22.0. All P values<0.05 were regarded as statistically significant.

Risk of MACE and low ATM did not show significantly different association; the same result was still observed after adjustment for potential confounders (Table 3, Figs. 2, and S5–S8).

Among the 1616 patients undergoing PCI only, 1598 (98.9%), 1570 (97.2%), 1601 (99.1%), 1168 (72.3%), and 1239 (76.7%) patients had filled prescriptions for aspirin, P2Y12 antagonists, statins, ACEIs/ARBs, and b-blockers, respectively, at discharge (Fig.S9). Among patients with at least one prescription filled, ATM of the PCI population and drug types of each medication class were determined, as shown in Fig.S10 and S11, respectively. Tables 2 and S1–S9 present the baseline characteristics of PCI population, CABG population, and conservative treatment population. In conservative treatment population, adherence to P2Y12 antagonists was significantly associated with age of patients (P<0.05).

Table 3 and Figs. 3 and S12–S15 show that in patients undergoing PCI, only low adherence to P2Y12 antagonists was significantly associated with higher risk of MACE (P = 0.029). The difference remained significant after adjustment for potential confounders (HR, 0.172, 95% confidence interval: 0.039 to 0.763; P = 0.021).

To measure medication adherence, some tools, such as self-reported Morisky Score, were applied. However, considering that self-reported Morisky Score is limited in settings with low rate of nonadherence [11], we opted to use PDC to measure medication adherence.

In the present study, we noted that patients who underwent PCI were generally more prone to refilling their prescriptions for P2Y12 antagonists during the first year after discharge than patients with CABG. This result is similar to those described by Hlatky et al. [12]. Similarly, patients treated using conservative methods were more vulnerable to discontinue taking aspirin, P2Y12 antagonists, statins, or b-blockers; possibly, in non-PCI-treatments, preventive medications are less probable to be prescribed, and patients are encouraged to undergo subsequent follow-ups, whereas after PCI, patients are educated regarding the importance of medications, particularly DAPT, in maintaining stent thrombosis [12]. Furthermore, patients without revascularization might not regard their conditions as serious and hence think that some medications are no longer necessary. Khanderia et al. discovered that patients undergoing CABG had concerns about overuse and harms caused by prescribed drugs, and whether low ATM was correlated with social, economic, and demographic factors [13]. On the other hand, patients undergoing CABG accounted for a very small proportion of our study group, indicating that results should be used with caution. Further research is needed to understand the causes of low adherence in non-PCI-treated patients and to develop appropriate interventions to improve ATM.

Previous studies demonstrated that co-payments also impact adherence [14,15], whereas in the present study, ATM is not associated with medical insurance, which might be attributed to the complex procedure of reimbursement and relatively high co-payment. Absence of statistically significant association was possibly caused by insufficient power, with demographic information on medical insurance available in only half of the population. Nonetheless, all these studies suggest that more effort should be given to improve medical insurance in China.

Regimen complexity also contributes to ATM. Prior studies [16,17] mainly emphasized the dosing frequency and multi-drug regimens, whereas ours considered different drug types under each medication class (for example, ticagrelor and clopidogrel of P2Y12 antagonists) and revealed that the mainly used drugs under each medication class were not associated with ATM. This result possibly indicates that using preventive medications is not limited by different drug types when considering ATM only. Furthermore, we explored the associations between ATM and multi-drug regimens and observed that adherence to statins, ACEIs/ARBs, and b-blockers improved with aspirin application. Similar results were observed with P2Y12 antagonists, indicating the improvement in adherence to statins. Interestingly, in adjusted analysis, multiple drug therapy did not affect ATM. Patients with concomitant use of aspirin and P2Y12 antagonists were more prone to undergo PCI, which was strongly associated with ATM. Results remained nonsignificant in PCI-treated and non-PCI-treated patients. Another study by Gislason et al. also revealed a similar result in which the multi-drug regimens during the study period did not worsen long-term adherence. However, by contrast, adherence to b-blockers and statins improved. The researchers also found that early initiation of multiple drug treatment during the study period was not associated with decreased long-term compliance. Their results suggest that noncompliance was not associated with increased efforts to administer more medicines in more patients during the study period [4]. Nonetheless, our study suggests that ATM associated with multi-drug regimens may not be as destructive as implied in previous reports [18,19].

As shown in many studies, low ATM, particularly that for DES thrombosis, which result from low adherence to aspirin and P2Y12 antagonists, is deleterious [20,21]. However, Kimura et al. doubted the actual role of P2Y12 antagonists and aspirin in preventing stent thrombosis; since then, more reservations were allotted for the effect of low adherence to dual antiplatelet medication. Likewise, in another study [7], low adherence to antiplatelet drugs was not associated with MACE (ACS and cardiac death), and result was explained by temporal interruption of antiplatelet medication and negligible rate of antiplatelet medication discontinuation during the first month after stenting; the first month is considered the most deleterious period [18]. Notably, association of absence of risk of MACE with low ATM was also observed in ACS patients in the present study. However, given the differences in definitions of MACE and ATM, these results should be interpreted with caution. In the present study, low ATM was defined as PDC<50%, and discontinuation of medications was defined as a break for at least three months. Nevertheless, in PCI-treated patients, the absence of risk for low ATM was generally consistent, with an exception for higher adherence to P2Y12 antagonists associated with lower risk of MACE, as approximated by the study of Airoldi et al. [19]; in that particular study, discontinuation of P2Y12 antagonists was used to determine occurrence of stent thrombosis within the first six months.

Several limitations must be acknowledged upon evaluation of results from this observational study. First, PDC of some patients was not accurately assessed because of the vague interval from index date to the exact day of medication discontinuation, that is, the available data were only the time at which medication was discontinued at each time point of follow-up (1, 6, and 12 months after discharge). We arbitrarily classified the patients into low adherence (PDC<50%) and high adherence (PDC>50%) groups; most studies defined high adherence as PDC>80%. Second, the relatively low rate of events after discharge in low ATM patients weakened the power to make a strong statement regarding the risk of MACE. Furthermore, a potential recall bias possibly existed in follow-up information, i.e., repeated calls probably caused patients to be more adherent to medications, and hence the estimated rate of ATM could be lower than that in other contexts. Third, the present study lacked objective measures for medication use. Testing the urine or blood to confirm successful medication would yield better results. Furthermore, the present study is a single-center study, which might have also caused certain bias. Finally, confounding bias was also possible.