The impact of statin discontinuation and restarting rates on the optimal time to initiate statins and on the number of cardiovascular events prevented

A patient is eligible for statins in England if they have a 10‐year risk of cardiovascular disease >10%. We hypothesize that if statin discontinuation rates are high it may be better to delay statin initiation until patients are at a higher risk, to maximize the benefit of the drug.


| INTRODUCTION
Cardiovascular disease (CVD) is the number one cause of death globally accounting for 31% of all deaths in 2017, 1 and contributes more than any other disease to the total disease burden around the globe. 2 Treatment for primary prevention of CVD is centered around lifestyle modifications such as changes to diet and exercise, and cholesterol-lowering medication such as hydroxymethylglutaryl-coenzyme reductase inhibitors (statins). There has recently been a lot of debate in the literature over the optimal risk to initiate statins for the primary prevention of CVD. Both England 3 (National Institute for Health and Care Excellence guidelines) and the United States 4 (American College of Cardiology/American Heart Association guidelines) have recently dropped their thresholds to a 10-year risk of 10% and 7.5%. However, the European Society of Cardiology still recommend a 10-year risk of a fatal CVD event of 5%, which equates to about a 15% risk of any CVD event, 5 while in Scotland the recommended threshold is 20% for asymptomatic individuals. 6 In support of higher thresholds, a recent study found that statins only provide a net benefit over possible harms at higher 10-year risks than the thresholds in current guidelines, and the benefits vary considerably by age and sex. 7 One factor that will affect the real-world impact of these guidelines is the widely reported suboptimal long-term adherence to and discontinuation from statins. [8][9][10] Studies examining factors affecting adherence to statins report consistent relationships between nonadherence and female gender, ethnic minority status, reduced income, lower number of concurrent CVD medications, new statin users, use of statins for primary prevention, smoking, depression, reduced follow-up and increased copayments, [11][12][13][14] while a recent high profile meta-analysis concluded that exaggerated claims about side-effect rates with statin therapy may be responsible for its under-use among individuals at increased risk of CVD. 15 The analyses underpinning the treatment thresholds do not incorporate the effects of non-adherence or discontinuation directly. We suggest that policy decisions around lowering of treatment thresholds may need to take account of real-world statin discontinuation rates in primary prevention. The reason could be that patients are initiating statins at a low risk and then discontinuing the drug when at a higher risk (risk increases with age), not maximizing the benefit of the drug.
The overall aim of this study was to assess the optimal time to initiate statins after a risk assessment in order to prevent the highest number of CVD events, given a patient's risk profile, and long-term adherence levels derived from real life data. We refer to adherence throughout this study specifically in relation to the combination of discontinuation and restarting rates. We also developed a range of scenarios where discontinuation rates were artificially decreased, allowing us to evaluate the effect that improving adherence would have on the number of CVD events prevented.

| Overview of simulation model design
A four-state health state transition model with cycle length of 1 year was created ( Figure 1) to answer our primary aim. Each scenario (age, gender, 10-year CVD risk score and assumed adherence rate) represented a patient having their 10-year risk assessment, which is F I G U R E 1 Design of the Health State Transition Model: p_c is the probability of a cardiovascular event; p_(c-adj) is the probability of a cardiovascular event, while receiving statin treatment; p_de is the probability of death (mortality); p_di is the probability of discontinuing statin treatment; p_r is the probability of restarting statin treatment [Colour figure can be viewed at wileyonlinelibrary.com]

KEY POINTS
• Current research evaluates when it becomes cost-effective to initiate statins for primary prevention of cardiovascular disease, but does not consider when is the best time in a patient's life to initiate treatment to prevent the most cardiovascular events • We hypothesised if discontinuation rates are high, it may be better to initiate statins at higher cardiovascular risks than 10% to ensure patients get the most benefit from the drug.
• Our results showed in some scenarios that there would be a benefit to delaying statin initiation past the 10% threshold, although the optimal time to initiate was based on age rather than cardiovascular risk • Such an approach has ethical concerns as you must base this decision on population level discontinuation rates, rather than the discontinuation time of the individual patient • Improving adherence in a meaningful way would lead to more cardiovascular events prevented than adjusting the threshold, but may be difficult to achieve when a clinician would decide whether to initiate statin treatment.
We varied the year of follow-up in which statins were initiated, and calculated the total number of CVD events expected. For the main analysis the discontinuation and restarting rates were derived directly from the Clinical Practice Research Datalink (CPRD) cohort, then for subsequent analyses the discontinuation rate were artificially decreased. The cost effectiveness of statins at various risk thresholds has already been extensively covered). 16 Instead, this model was set up to calculate the number of incident CVD events prevented by initiating statins at different times, assuming real life risk profiles and adherence rates, and is what makes this study unique.

| Data source
This project used data from the CPRD linked with Hospital Episodes Statistics (HES) and Office for National Statistics (ONS). CPRD is a primary care database representative of the United Kingdom in terms of age, sex and ethnicity, 17 although linkage to HES restricts this dataset to England only. The data were used to create two cohorts, a cohort of statin users (statin cohort) and a cohort of patients at risk of CVD (primary prevention cohort).
The primary prevention cohort was defined in the same way as the QRISK3 development cohort. 18 To be eligible for the cohort, a patient must have had 1 day of follow up in CPRD that met the following inclu- to ensure that all patients were first time users of statins, rather than current users who have transferred from another practice. A patient entered the cohort on the date of their first statin prescription and exited the cohort at the end of that statin treatment period (detailed definition in Appendix S2). A patient could leave and re-join the cohort (at the start of their next treatment period) multiple times before their censoring date. A patient was censored if transferred out of practice, at the end of data collection, death or occurrence of a CVD event.

| Estimation of transition probabilities
CVD event probabilities were calculated from the primary prevention cohort. A lifetime risk model was fitted using standard techniques for developing lifetime risk models. 19 Kaplan Meier curves were then fit to each group and the probability of a patient discontinuing/restarting during each day of follow up was calculated. As the duration of follow-up in the simulation was longer than in our data, the discontinuation/ restarting rates were extrapolated. If a patient discontinued for a third time we made the assumption they did not restart treatment because the discontinuation rate in the fourth treatment period was high (76%/90% after 1/2 years), and only 314 patients remained in this cohort after 5 years (see Section 3).
For the first treatment period discontinuation rates were stratified by age (this was not possible for subsequent periods as sample size was deemed too small for some subgroups). A Cox proportional hazards model was fit to the discontinuation data from the first treatment period with age as a predictor variable, considering fractional polynomials of age using the mfp package. 22 This allowed the discontinuation rate to be a function of age. Full details of the stratification and extrapolation of the discontinuation rates is provided in Appendix S2.
The transition probabilities of non-CVD related death were calculated using the primary prevention cohort. The date of death was based on the data as recorded in primary care, shown to have 92% concordance with ONS within 2 weeks. 23 These data were combined with ONS, for which we had linkage to CVD related deaths. Deaths identified in primary care that were CVD related were then excluded. Incidence rates of death across each age category were then calculated.

| Implementation of the simulation
Different scenarios were simulated based on a patient having a risk assessment (start of the simulation), and the decision of whether to initiate statins straight away, or delay. Variables that made up the different scenarios were: age, gender and 10-year CVD risk at the start of the simulation, the statin initiation date, and an assumed adherence rate. The ages considered were 40, 50 and 60. For each age, we considered all 10-year risks within the 1 to 99th percentile range of risk scores calculated for patients in that age group from our primary prevention cohort.
The statin initiation date was varied in yearly intervals from the start of simulation. Given the discontinuation rate for the first treatment period was stratified by age, this meant the age at statin initiation time impacted the discontinuation rate used in each scenario. Duration of follow up was from the age at start of the simulation (risk assessment), until 90 years of age. Cycle lengths were 1 year. For each scenario we simulated 10 000 patients and calculated the number of CVD events over the course of the entire duration of follow up, which was compared with the number of events if no statins were given, providing the number of events prevented per 100 people.
This process was repeated using four different adherence rates.
The discontinuation rate from the first, second and third treatment periods were altered so that the probability of discontinuation was 5/6th, 2/3rd, ½ or 0th (100% adherence) of the rate derived from CPRD.

| Sensitivity analyses
The simulation was conducted assuming a treatment effect of 0.65 and 0.6, given the uncertain nature of the estimate used in the primary simulation. Also, simulations were conducted using discontinuation and restarting rates from a cohort of statin users where any Baseline   Of all the patients that discontinue, 50% have restarted a year after the initial discontinuation, 59% after 2 years, and 79% after 10 years.

| Discontinuation and restarting of statins
The second discontinuation and restarting rates suggest patients are more likely to discontinue/restart during the subsequent treatment periods. Graphs for the discontinuation rate in the first treatment period stratified by age, extrapolated discontinuation and restarting rates beyond our period of data, and discontinuation and restarting rates for the cohort of long term statin users (no single prescriptions), are all presented in Appendix S2.  Figure 4 shows the effect of reducing the discontinuation rate to 5/6, 2/3, 1/2 of the rate we found in practice, and no discontinuation. For each age group, a single 10-year CVD risk (close to the median of that age group) was selected to showcase the effects, so all trajectories within a plot consider the same group of patients. It

| Effect of increasing statin adherence on CVD events
shows the more adherent to statins people are, the more benefit they receive, and this benefit is increased the earlier prescribing is initiated (for males; results for females are presented in Appendix S3). Results from all sensitivity analyses outlined in the methods are provided in Appendix S3. A small discussion is also provided, the results echoing those from the primary analysis.

| DISCUSSION
There are three key findings from this study. The first is that between the ages of 40 to 70, the statin initiation time had a meaningful effect on the number of events prevented. Furthermore, the risk score of a patient had a negligible effect on the optimal time to initiate statins, F I G U R E 3 Number of cardiovascular disease (CVD) events prevented over the duration of follow up with different time delays in starting statins, stratified by baseline age and 10-year risk of cardiovascular disease, using the discontinuation rates as observed in the statin cohort (males) [Colour figure can be viewed at wileyonlinelibrary.com] which was driven by age. The second is that discontinuation and restarting rates get higher with consecutive treatment periods, underlining a complex pattern of statin usage over time. The third is that large gains could be made by improving adherence.
We observed fairly large differences in the number of events prevented when statins were initiated between the ages of 40 to 70 with a peak around ages 59 (male) and 63 (female), regardless of the CVD risk scores of the patients. Initiating statins below the age of 50 was associated with far fewer events prevented, however it is unlikely for patients this young to have a CVD risk of 10% (the threshold for cost effectiveness), and so this is unlikely to happen in practice. However it is not uncommon for a 50-year old to have a CVD risk of 10%. Our data indicates that delaying statin initiation by 10 years could prevent an extra 0.67 events per 100 men treated, and 0.96 events per 100 women treated. These gains are small but not insignificant, and are likely to be driven by the observation that adherence improves with age (until around age 70, Appendix S2: Supplementary Figure 2.3), and that patients who restart statins for the second time or more are less likely to continue with treatment ( Figure 2). There is therefore a middle ground to be found which ensures patients are offered the drug when they are most adherent, at a high enough risk to gain benefit, but also that the risk of death or having a CVD event prior to receiving treatment is small enough.
Interestingly, for a given adherence level, the optimal time to prescribe was driven primarily by age rather than the 10-year CVD risk.
In Figure 3 the maxima of each trajectory are at the same age despite differing risk levels. This suggests that given the suboptimal adherence levels in practice, in order to prevent the most events in the population, the optimal time to initiate statins for men is around 59 (women 63), irrespective of the CVD risk score of the patient.
While the CVD risk score may drive the cost-effectiveness of statin treatment, it does not drive the optimal time in a patient's life to start taking statins to prevent the most CVD events, which our work suggests is driven by age. The distinction can be highlighted by if a patient has perfect adherence (Figure 4), the optimal time to initiate statins is as early as possible, but the treatment may not be cost effective at this point.
The potential to prevent more CVD events in the population using such an approach brings up some important ethical concerns.
Gains would be made from ensuring that all patients will receive the drug when it will have most benefit (not too early, not too late). However, alongside any gains made by delaying statin initiation to a certain age, there will be a cost to adherent patients who would have continued treatment if starting at a younger age. Arguably it is unethical to improve the health of the population in this manner. In an ideal world we would know the adherence of a patient before initiating them on treatment, and could then initiate at the most appropriate time for that patient. Unfortunately this is not possible, and we would be forced to use population level discontinuation rates, which has these concerns.
We found inconsistent use of statins by patients in primary prevention. We also found higher discontinuation and restarting rates during the later treatment periods. This provides extra information beyond the current literature, which reports the initial discontinuation and restarting rates. 10 The present study found that improving adherence could have a larger impact than adjusting statin initiation thresholds. This is not unsurprising, given this results in more time on treatment, however could be difficult to achieve. The most recent  24 Like other studies, 10,25 this study suggests that people are likely to discontinue their statin when it is newly prescribed. Targeting a patient-centered, theorybased low-cost intervention which focuses on patients' concerns during this key initial period has been shown to improve adherence by 11% in a range of chronic illnesses, 26,27 and forms the basis of a National Health Service commissioned service in England (New Medicines Service 28 ). This service is not currently provided to people starting statins, however, a randomised controlled trial of delivery of the same intervention in long term statins users demonstrated improved adherence. 29 This suggests that extension of the New Medicine Service into statin users could demonstrate effectiveness.
There were three key limitations we identified in this study.
(a) We used prescription data as a proxy for patients taking statins.
This is a limitation as we only know a patient was given a prescription by their GP, we do not know if they picked the drug up, or took the drug. Therefore there is a possibility discontinuation rates are higher in practice, which would push the optimal time to prescribe further back. However there is currently no better way to measure adherence in the United Kingdom on a large scale, until prescribing and dispensing data are linked. Secondly, we only consider patients on treatment if they continually pick up their prescriptions (ie, our algorithm). We think it is unlikely patients will have discontinued treatment but continue to pick it up. (b) We extrapolated the statin discontinuation and restarting rates for the length of the simulation. Data on statin usage over an individual patient's lifetime would be highly valuable to inform work such as this, but is not available. (c) We did not stratify the second and third discontinuation rates or first and second restarting rates based on age, despite age being a predictor of statin adherence. 30 Our reasoning is that this would have significantly reduced the cohort size available to calculate discontinuation rates, a particular issue for the second and third treatment periods at 10 years follow up. Given we were extrapolating data from this point, this was undesirable. Given the impact of discontinuation rates on the optimal time to initiate therapy, further work could be done to explore the impact of changes in statin intensity and dose on discontinuation rates, and subsequently the best time to implement these changes.

| Conclusion
In certain scenarios, extra CVD events could be prevented by delaying statin initiation beyond a risk of 10% until reaching a certain age (59 for men, 63 for women). These findings are based on the discontinuation and restarting rates in England. Currently all thresholds are based around a patient's risk score, which drives cost effectiveness.
However a combination of age and adherence levels are the most important factors in determining the optimal point in a patient's life to initiate statins. However, the clinical benefit must be weighed up against ethical concerns of such a strategy. We cannot know when a given patient will discontinue treatment in advance, and using population level discontinuation rates may disadvantage the most adherent patients. A less controversial strategy which could result in preventing more events would be to focus on improving adherence, although this may be harder to achieve.

ETHICS STATEMENT
This study is based in part on data from the Clinical Practice Research Datalink obtained under licence from the UK Medicines and Healthcare products Regulatory Agency. The study was approved by