OUP user menu

★ editor's choice ★

Discontinuation of statin therapy following an acute myocardial infarction: a population-based study

Stella S. Daskalopoulou , Joseph A.C. Delaney , Kristian B. Filion , James M. Brophy , Nancy E. Mayo , Samy Suissa
DOI: http://dx.doi.org/10.1093/eurheartj/ehn346 2083-2091 First published online: 29 July 2008

Abstract

Aims Randomized clinical trials have shown that statins can reduce mortality after acute myocardial infarction (AMI). However, the impact of changes in patterns of statin use, particularly stopping statins, on survival post-AMI is unknown. Our objective was to estimate the extent to which different patterns of statin use are associated with post-AMI mortality.

Methods and results Population-based, cohort study, from 2002 through 2004 in the United Kingdom General Practice Research Database (GPRD), involving patients surviving 90 days after their first AMI. Past statin use was defined as any statin prescription within 90 days before AMI; statin use post-AMI as any statin prescription within 90 days after AMI. Cohort entry was at day 90 post-AMI; subjects were followed for 1 year. Four groups were identified: (i) non-users (patients never on statins); (ii) users (on statins before and continued post-AMI); (iii) starters (started statins after the event); and (iv) stoppers (stopped statins after the event). Hazard ratios (HRs) were estimated using Cox proportional hazards model. The main outcome measure was 1-year all-cause mortality. The cohort included 9939 AMI survivors (mean age: 68.4 ± 12.8 years; 60.3% men), 22.7% of whom were not prescribed a statin post-AMI. When the non-user group (n = 2124) was considered as the reference, the adjusted HRs (95% confidence intervals) of death were 0.84 (0.66–1.09) for users (n = 2026), 0.72 (0.57–0.90) for starters (n = 5652), and 1.88 (1.13–3.07) for stoppers (n = 137). Stoppers of control medications (aspirin, β-blockers, and proton pump inhibitors) were not associated with increased mortality.

Conclusion Discontinuation of statins in survivors of a first AMI was relatively rare in this cohort. However, statin discontinuation was associated with higher total mortality and this may represent a biological rebound or/and a risk-treatment mismatch phenomenon, where treatment is withdrawn from very ill patients. While awaiting further research, at present statin use should only be withdrawn under judicious clinical supervision.

Keywords
  • Acute myocardial infarction
  • Statins
  • Discontinuation
  • Biological rebound phenomenon
  • Risk-treatment mismatch
  • General Practice Research Database
See page 2061 for the editorial comment on this article (doi:10.1093/eurheartj/ehn348)

Introduction

Coronary heart disease (CHD) is the leading cause of morbidity and mortality worldwide.1 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors (also known as statins) have been shown not only to lower cholesterol levels but also to reduce cardiovascular morbidity and mortality. A meta-analysis of 90 056 individuals in 14 primary and secondary prevention randomized statin trials have demonstrated a 19% relative reduction in coronary mortality [relative risk (RR), 95% confidence interval (CI): 0.81, 0.76–0.85; P < 0.0001] during a mean of 5 years.2 Significant reductions were also noted in acute myocardial infarction (AMI) or coronary death, the need for coronary revascularization, fatal or non-fatal stroke, and a reduction of 21% in a composite endpoint of all the above major vascular events.2 Based on these findings, treatment guidelines strongly endorse cholesterol-lowering medications.35

However, despite the beneficial effects of statins and the relatively rare serious side effects, statin use is variable in actual clinical practice. The discontinuation rates for use of this usually life-long treatment are high, being approximately 30% within the first year of prescription, and are broadly similar with all statin drugs.6,7 The drug discontinuation rates in routine care are significantly higher than those observed in clinical trials.6,7 Previous studies have shown that statin discontinuation post-AMI has a short-term harmful effect on survival. However, less is known about the impact of changes in patterns of statin use, particularly that of stopping statins, in the general population on longer term (1 year) mortality post-AMI. Therefore, we sought to estimate the association between 1-year all-cause mortality following AMI and different patterns of statin use.

Methods

Study design

We conducted a population-based, cohort study of patients who survived a first AMI between 1 January 2002 and 31 December 2004 in the UK. Data were extracted from the UK General Practice Research Database (GPRD), obtained from the UK Medicines and Healthcare Products Regulatory Agency. This database has been described in detail elsewhere.8,9 Briefly, the GPRD is a large clinical database based on information generated from general practices in the UK; more than three million people are enrolled with more than 400 general practitioners. This database is based on the computer records of a network of general practitioners who enter the data on their patients using a standard protocol. Importantly, the age and sex distribution of patients in the GPRD is representative of the population of England and Wales.8 The GPRD also includes information on demographic and lifestyle variables, such as height and weight, smoking status, and alcohol consumption.8 This type of information is often unavailable in other population prescription claims databases.9 Furthermore, drug prescriptions are recorded in GPRD in detail using a coded drug dictionary based on the UK Prescription Pricing Authority. Medical diagnoses are entered based on a modification of the Oxford Medical Information System classification. The accuracy and completeness of GPRD data have been previously validated (including for AMI) and proven to be of high quality.810 The GPRD has been used in over 500 peer-reviewed publications, including several studies on AMI.

To be eligible for cohort entry, patients needed to survive at least 90 days following their first AMI. Also, patients had to be at least 20 years of age and have a minimum of three consecutive years of records in the GPRD before the AMI to be eligible for inclusion in our study. This criterion was used to ensure that all patients had an adequate time to be seen by their general practitioner in order that their medical history and comorbid conditions would be recorded in the GPRD. This criterion also helped ensure that our cohort was restricted to patients with a first AMI.

The study was approved by the Scientific and Ethical Advisory Group of the GPRD, and the ethics review board of the McGill University Health Centre.

Exposure definition

Past statin use was defined as any statin prescription in the 90 days before AMI; statin use post-AMI was defined as any statin prescription in the 90 days after AMI. Patients who died within 90 days after their AMI were excluded from the study. Cohort entry (time zero) was, therefore, day 90 post-AMI. These criteria were introduced to allow equal opportunity for all patients to receive a statin prescription across a clinically relevant time period thereby avoiding immortal time bias (an artificial survival advantage associated with the exposed group regardless of the effectiveness of the treatment).11,12 The 90-day exposure time window is widely used in pharmaco-epidemiological studies (including many studies in the GPRD) as the longest length of prescription that can be issued in the UK is 90 days. Thus, the use of a 90-day exposure time window minimizes the potential misclassification of exposed patients as unexposed. This approach to cohort design and analysis produces valid estimates of the hazard ratios (HRs).11,12

We identified four groups of patients according to the pattern of their statin use: (i) non-users, consisting of patients not receiving statins before or after AMI; (ii) users, consisting of patients receiving statins before and continuing after AMI; (iii) starters, consisting of patients who did not receive statins before their AMI but who started after it; and (iv) stoppers, consisting of patients who stopped statin therapy after their AMI.

Outcome definition

The outcome of this study was all-cause mortality between 90 days and 1 year after the AMI. Subjects were followed until the occurrence of the outcome (death) or the end of the study (1 year after the AMI or 270 days following cohort entry), whichever came first.

Covariates

Covariates included age, sex, smoking status (as a binary variable: never/ever smoker), alcohol abuse (by clinical diagnosis), obesity (defined as body mass index ≥30 kg/m2), and number of hospitalizations in the past year before the AMI. Comorbid conditions before the AMI were identified from physicians' diagnoses (presence or absence) any time before the AMI and prescriptions issued in the 90 days before the AMI (Table 1). There were 8 and 20% missing data for smoking and obesity, respectively.13

View this table:
Table 1

Baseline characteristics of all patients and of the four groups of patients according to the pattern of their statin use during the pre- and post-acute myocardial infarction (AMI) periods

Statistical analysis

Descriptive statistics were performed to describe baseline characteristics of the study cohort. HR were estimated using Cox proportional hazards model with corresponding 95% CIs. The proportional hazard assumption was verified using the graph of the log(-log(survival)) vs. log of survival time graph and visually assessing that the lines were approximately parallel. We assessed the effect of discontinuing statins post-AMI using a crude unadjusted model and an adjusted model generated with stepwise variable selection. All variables identified in Table 1, as well as post-AMI medications, were considered for possible inclusion into the adjusted model. While most variables were dichotomous, age and number of previous hospitalizations were continuous and were assessed graphically to determine if there were violations of the linearity assumption. There was no improvement in model fit when introducing non-linear transformations of these variables. Consequently, only the linear covariates were included in the final model for parsimony. We performed an ‘intention-to-treat’ analysis, with medication use being assessed in the 90 days between AMI and cohort entry. According to this analysis, persons who were prescribed statin therapy were considered to always be taking statins throughout the follow-up period. Similarly, those who were not prescribed statins during this assessment window were considered to never take this medication throughout the follow-up period.

To test the validity of this approach, we included three indicator variables for proton pump inhibitors (PPIs) treatment patterns (users, starters, and stoppers). PPIs thus served as a ‘control drug’, which enabled us to assess the effect of different patterns of drug prescription on a drug that is not prescribed for cardiac diseases. We extended our analysis to also include aspirin and β-blockers; this extension aimed to assess whether the effect of withdrawal was associated with discontinuation of any cardio-protective medication or was restricted to statins. Indicator variables for the different patterns of use of these medications were also included in our Cox proportional hazards model. We also evaluated the tendency to stop, continue or start other cardio-protective medications in the four identified groups of patients to ensure that any effect on mortality was due to stopping statins and not a pattern of stopping many medications simultaneously. All analyses were performed using SAS software, version 9.1.3 (SAS Institute Inc., Cary, NC), and a two-tailed P < 0.05 was considered statistically significant.

Results

The cohort included 9939 survivors of their first AMI (mean age ± SD: 68.4 ± 12.8 years; 60.3% men). Of those patients, 2163 (21.8%) had a statin prescription in the 90 days before their AMI, while 7678 (77.3%) were prescribed a statin within 90 days after AMI. From the 2261 patients who were not prescribed a statin post-AMI, 2124 were non-users (neither before, nor after the AMI) and 137 were on a statin before the AMI and stopped this medication after the event. The follow-up of the patients was excellent (only 1.75% of the subjects were lost to follow-up).

Table 1 shows the baseline demographic, lifestyle and clinical characteristics, including comorbidities and medication used of all AMI survivors. Baseline characteristics are also stratified according to the pattern of their statin use during the pre- and post-AMI periods. Known predictors of poor prognosis, including age (adjusted HR, 95% CI: 1.05, 1.04–1.06), smoking (adjusted HR, 95% CI: 1.49, 1.24–1.78), congestive heart failure (adjusted HR, 95% CI: 1.60, 1.28–1.99), renal failure (adjusted HR, 95% CI: 1.92, 1.34–2.78), and diabetes (adjusted HR, 95% CI: 1.34, 1.09–1.66) were identified under multivariable analysis as statistically significant predictors of all-cause mortality between 90 and 365 days after the AMI.

Kaplan–Meier analysis suggests a significant difference in time to mortality between treatment groups (P < 0.0001) (Figure 1). Similarly, unadjusted Cox proportional hazards models suggest that, when the non-user of statins group (n = 2124) was used as the reference group, a protective effect of statins was found in the user and the starter groups (Table 2). In contrast, stopping the statin had a harmful effect on survival (HR, 95% CI: 1.60, 1.05–2.43). However, the crude model for PPIs showed that there was a harmful effect for both the stopper and the user of PPIs groups, while there was no effect of starting PPIs. Similar results were found for aspirin and β-blockers crude estimates (Table 2).

Figure 1

Kaplan–Meier analysis of effect of statin treatment patterns on 1-year all-cause mortality among survivors of a first acute myocardial infarction (AMI). Non-users were defined as patients not receiving statins before or after AMI; users those receiving statins before and continuing after AMI; starters those who did not receive statins before their AMI but who started after it; and, stoppers were defined as those who stopped statin therapy after their AMI.

View this table:
Table 2

The effect of different prescription patterns for statins, proton pump inhibitors (PPI), aspirin and β-blockers on the hazard ratio (HR) for 90-day to 1-year all-cause mortality among the survivors of a first AMI

After adjustment for baseline differences and when the non-user group (for statins and PPIs as well as aspirin and β-blockers) was used as the reference group, a strong harmful effect of statin discontinuation after the AMI was shown (adjusted HR, 95% CI: 1.88, 1.13–3.07) (Table 2). For the ‘control drugs’ (PPIs plus aspirin and β-blockers), none of the different patterns of use had a statistically significant increase on the 1-year mortality of AMI survivors despite a trend among aspirin stoppers (Table 2).

Patients in the statin stopper group had higher rates of discontinuation of pre-AMI cardio-protective medications than patients in other groups (Table 3).

View this table:
Table 3

The pre- and post-AMI prescription patterns of patients by statin use category; percentage of patients who were prescribed a cardio-protective medication in the 90 days before and 90 days after first AMI recorded in the General Practice Research Database

Discussion

Using a large, population-based cohort, we demonstrated that discontinuation of statin therapy after an AMI is associated with a higher mortality than any other pattern of statin prescription. This finding builds on recent work that demonstrates a positive relationship between adherence to statins and survival after AMI in a graded dose–response type fashion.14 However, to our knowledge, this is the first study to show in actual clinical practice that patients who discontinuing statin treatment post-AMI have a significantly higher long-term risk than statin non-users.

Statins are known to confer clinical benefits which extend beyond their lipid-lowering effects.15 Thus, besides reducing cholesterol biosynthesis, inhibition of mevalonate by statins also leads to a reduction in the synthesis of important intermediates of the cholesterol biosynthetic pathway that modulate a variety of cellular processes including cellular signalling, differentiation, and proliferation.16 Statins, through reduction in the synthesis of these intermediates, improve vascular function exerting the so-called ‘pleiotropic effects’.15

Evidence from clinical studies shows that the beneficial effects of statins are rapidly lost and often transiently reversed when statins are acutely discontinued.16 This might be more profound following an acute coronary event, rather than in stable chronic vascular disease. Data from the Treating New Targets (TNT) study suggested that short-term discontinuation of statin therapy in patients with stable cardiac conditions may not substantially increase the risk of acute coronary syndromes (ACS).17 On the other hand, in a study that used data from the National Registry of Myocardial Infarction 4, the effects of early (within 24 h) statin treatment of AMI patients on in-hospital mortality were evaluated. When compared with patients who were not treated with statins, those who continued or newly started statin treatment had a decreased risk of in-hospital mortality; however, discontinuation of statins was associated with a higher risk of in-hospital mortality (compared with no statins).18 In patients with non-ST elevation, AMI discontinuation in the first 24 h was associated with a 2-fold risk when compared with continuation of statin treatment.19 However, neither of these studies examined our question of the long-term (1 year) impact of statin discontinuation.

Statins are shown to reduce inflammatory markers.20 However, it is relevant that statin withdrawal can result in increase in C-reactive protein levels, as reported in a study of healthy individuals with hyperlipidaemia.21 The increase to pretreatment levels occurred on the second day after withdrawal of statins suggesting that the anti-inflammatory effects of statins are rapidly lost after treatment withdrawal; these effects were unrelated to low density lipoprotein cholesterol levels.21

Although there is evidence supporting a biological rebound phenomenon after statin withdrawal, another possibility that could explain our findings is a risk-treatment mismatch. The discontinuation of statin use could occur in patients with a higher probability of death.22,23 Information on the reasons of statin discontinuation is not available to us; however, no obvious reason for statin withdrawal was identified in the patients' profile, but we cannot exclude the possibility of unknown confounders. In fact, the stoppers had a very similar risk factor profile with the users, and no potential contraindications for statin prescriptions or life-threatening conditions were identified. Therefore, these patients should benefit from effective, evidence-based treatments such as statin therapy.22,23 Other studies have also shown the paradoxical phenomenon that prescription of statins diminished progressively as baseline cardiovascular risk and future probability of death increased.22,23

In our study, although the vast majority of the patients were prescribed a statin after their AMI, it is relevant that >20% of patients were not offered such treatment. There were some important differences in baseline characteristics between groups, including non-users being older than patients in the other groups. Underevaluation and undertreatment appear to be common in the elderly.22,23 However, the Prospective Study of Pravastatin in the Elderly at Risk (PROSPER) and the Heart Protection Study (HPS) offer evidence that statins are as efficacious in elderly patients as in younger individuals.24,25 Patients who started statins post-AMI had lower mortality rates than non-users of statins. Protective associations were also observed for starters of aspirin and β-blockers (Table 2).

The observed HR of the stopper group for all-cause mortality post-AMI was 1.88 (95% CI: 1.13–3.07) which suggests a potentially important impact of discontinuation on mortality. To determine if the risk was solely due to prescription discontinuation post-AMI or was statin-specific, we included ‘control drugs’ in our model (PPIs plus aspirin and β-blockers). None of the different prescription patterns (for PPIs or other cardiovascular drugs) had any statistically significant deleterious effect on survival, which shows that the risk of higher mortality is not associated with discontinuing a drug per se. In addition, the statin stoppers did not show a strong tendency to discontinue aspirin and β-blockers meaning that discontinuation of these medications cannot fully explain the steep rise in risk of death among statin stoppers.

A recent study showed that discontinuation of statins after acute ischaemic stroke was associated with a higher 1-year all-cause mortality (adjusted HR, 95% CI: 2.78, 1.96–3.72).26 In this study, 38.9% of the patients discontinued their statin, in 71.2% of these cases there was no specific reason for discontinuation, whereas in the remaining 28.8% the reason was mild side effects.

Our study has both clinical and policy implications. Discontinuation of statins was associated with worse survival after the first AMI. Whether this is a biological rebound phenomenon, or risk-treatment mismatch, or both cannot be identified. However, patients after an AMI are at high risk of events (including death) and should be treated aggressively in the absence of clear contraindications. As the survival of statins stoppers was worse than the non-users, medical professionals should be very careful with drug discontinuation post-AMI. Physicians also need to increase awareness of patients regarding the implications of discontinuation of their medications (sometimes despite doctors' recommendations) and encourage higher adherence. A study by Jackevicius et al.27 using administrative data from Ontario evaluated 2-year adherence rates to statins among people older than 66 years of age with recent ACS, those with chronic CHD (secondary prevention), and those without CHD (primary prevention). The 2-year adherence rates were only 40.1% for ACS, 36.1% for chronic CHD, and 25.4% for primary prevention. Although patients with ACS had better adherence relative to the other groups, elderly patients with or without recent ACS had low rates of adherence to statins.27 It was suggested by the authors that many patients initiating statin therapy may receive no or limited benefit from statins because of premature discontinuation.

In our study, 21.8% of the patients were on statins before the event. This is in accordance with findings of other studies; the Platelet Receptor Inhibition in Ischemic Syndrome Management (PRISM) trial,28 the National Registry of Myocardial Infarction 4,19 the study from Ontario,27 and the multinational Global Registry of Acute Coronary Events (GRACE) study29 also found that only 15–30% of patients admitted for an acute coronary event were taking statins before their event.

In the PRISM trial,28 the effect of statins on cardiac events among patients who had coronary artery disease and chest pain in the previous 24 h was examined; 40% of patients were on statins pre-AMI. Statin withdrawal was associated with a substantially higher 30-day event compared with those who continued statin therapy (adjusted HR, 95% CI: 2.93, 1.64–6.27). The higher prevalence of pre-AMI statin use in the PRISM trial may be explained by the fact that patients participating in trials are generally not representative of the general population. The National Registry of Myocardial Infarction 4 showed that 25% of patients were taking statins prior to their non-ST segment AMI.19 Patients who discontinued statin therapy within 24 h of their AMI had a significantly higher risk of in-hospital death. Unlike these two previous studies, our study, which includes general population, examined longer term (1 year) all-cause mortality. The study by Jackevicius et al.,27 which used data from Ontario, also found that only 15–30% of patients admitted for an acute coronary event were taking statins before their event. This descriptive study did not evaluate the hazard of discontinuing statin therapy. Although it is also a population-based observational study,27 it is restricted to patients older than 66 years of age, whereas our study included all patients regardless of age. On the other hand, the GRACE study29 included all patients above 18 years of age who presented with ACS in 94 hospitals located in 14 countries. Unlike ours, this study evaluated outcomes (in-hospital events and deaths) within 7 days of the presenting event, and not longer term (1 year) mortality. It was shown that in this short period after the event (7 days) statin stoppers were as likely to die during hospitalization [odds ratio (OR), 95% CI: 1.39, 0.91–2.14] as non-users of statin therapy. However, patients who continued, or started statins in the hospital were less likely to experience complications or die than patients who never received statins (OR, 95% CI: 0.66, 0.56–0.77 and OR, 95% CI: 0.38, 0.30–0.48, respectively). These findings suggest that statin therapy can beneficially modulate early pathophysiological processes in patients with ACS.29

Our study has a number of strengths. First, it includes all patients with a first AMI giving a large sample size with excellent follow-up. Since we did not focus only on older patients as other population-based studies have done, our generalisability is expected to be greater. Unlike clinical trials, our study involves a representative sample of unselected subjects (both men and women) and reflects a real-world setting. Finally, the GPRD contains detailed clinical and lifestyle variables not typically available in prescription claims databases.

Our study has several limitations. First, we did a class-effect analysis of statins instead of looking at individual statins due to lack of power to divide statin stoppers. However, an earlier population-based 5-year study compared five statins using data from medical administrative databases in three provinces of Canada (Quebec, Ontario, and British Columbia) and found that statins were equally effective for secondary prevention in elderly patients after AMI.30 Also, there is evidence that physicians treat statins as a class and do not choose their statins based on clinical trial evidence.31 Second, information on the severity of AMI was not available to us. Third, we did not have information on in-hospital treatment. Fourth, we restricted our analysis to recent years (2002–2004) to reflect the therapeutic policy after the publication of the stringent guidelines for lipid lowering in patients with CHD.3 We did not include the latest years as statins became available over the counter in the UK and could have resulted in some minor exposure misclassification.32 Fifth, the reasons for discontinuation of statins are not available in the GPRD. Also we used issued prescriptions (information derived from electronic records) to estimate actual pill intake. However, this is a standard method used in population-based databases (and in the GPRD).810 Although a fully time-dependent measure would have given a more precise estimate of the effectiveness of the statin therapy, it would not have directly addressed the primary study question of the impact of changes in prescription patterns. For this research question, an ‘intention-to-treat’ (or, in our case, ‘intention-to-change therapy’) analysis is the more appropriate estimate. Sixth, as is true of most observational studies, we may not have been able to completely control for potential confounders related to severity of illness or excess comorbidities. Thus, although we adjusted for a number of important risk factors and potential confounders, our study may be affected by residual confounding. Possible residual confounding and lack of information about the severity of illness could at least partly explain why statin stoppers are also more likely to stop other medications than patients in the other groups. However, the GRACE study29 showed that patients who present with ACS and were on statins before the event had less severe presentation, fewer in-hospital complications, and lower hospital death rates than patients not on statins before the event.

In conclusion, discontinuation of statins in survivors of a first AMI was associated with higher all-cause mortality when compared with non-users. Whether this represents a biological rebound phenomenon or/and risk-treatment mismatch remains to be resolved. Regardless of mechanism, the medical professionals should be careful when reviewing patients' medications and the patients should have high adherence to effective cardio-protective medications. As the performance of randomized clinical trials examining the withdrawal of statins, an evidence-based drug, from patients after an AMI is not ethical, the evidence can only be extracted by large, observational population-based studies. Further population-based studies of statin discontinuation are needed to confirm our findings given their potentially important clinical implications.

Conflict of interest: none declared.

Funding

Dr Brophy and Mr Filion are supported by Les Fonds de la Recherche en Santé du Québec. Dr Suissa is the recipient of the CIHR Distinguished Scientist Award. Dr Suissa reported that he is currently or in the past 5 years has been a consultant or advisory board member for or has received grants from AstraZeneca, Bayer, Boehringer-Ingelheim, Bristol-Myers Squibb, GlaxoSmithKline, Merck, Organon, Ortho, Pfizer, Sanofi-Aventis, Schering, Schering-Plough, Sepracor, and Wyeth. Mr Filion reported that, in the past 5 years, he was a consultant for Isaix Technologies, Inc.

References

References

The above article uses a new reference style being piloted by the EHJ that shall soon be used for all articles.

View Abstract