In this modeling study, we found that in 2,428 asymptomatic participants, statin therapy resulted in robust, small gains in total life expectancy and somewhat larger gains in CHD/stroke-free life expectancy. The expected benefit of statin therapy was determined by a number of baseline variables. From these variables, we constructed a web-based calculator and color charts. Once the underlying model has been independently validated, these tools can be used for communication of the expected lifetime benefits of statin therapy in persons aged 55 y and older. Inconsistencies occurred between the predicted benefits and what can be expected from the currently recommended 10-y CVD risk assessment. These inconsistencies were predominantly caused by age, which acts on lifetime benefits in the opposite direction as its effect on 10-y CVD risk. Individuals at low 10-y CVD risk may achieve a similar or even larger gain in total and CHD/stroke-free life expectancy as those at high 10-y risk.
Most decision tools for CVD prevention in asymptomatic individuals predict the individual's risk over a time period ranging from 5 to 10 y without calculating potential treatment benefits 
. If treatment benefits are presented, they are usually calculated as absolute risk reductions without taking into account competing risks 
. Two decision tools for making choices on statin therapy were based on Markov models predicting lifetime outcomes with and without statin therapy 
. The underlying decision models used data from multiple sources for estimating CVD events, and age- and sex-specific life tables for competing death probabilities, which are not necessarily compatible 
. In contrast, we used event probability estimations from one data source. Furthermore, we modeled the occurrence of stroke events separately from CHD events. Statin therapy has a different effect on strokes 
, and ignoring this effect would lead to incomplete estimation and communication of treatment benefits.
Despite these strengths, our results must be interpreted in the light of some limitations. One limitation is that the RISC model was used to extrapolate 5-y predictions to a lifetime horizon, which may be very sensitive to the method chosen 
. The RISC model extends cumulative incidence functions by updating age and risk factor levels using 5-y time intervals. Secular trends in risk factor levels were modeled across the age span using cross-sectional data, and thus potential chronological and cohort effects were not taken into account. We evaluated the validity of these extrapolations with subsequently available Rotterdam Study data not used in developing the RISC model and found that the deviations were generally limited. Developing predictions on longer follow-up data, e.g., 30 y, would allow for a more comprehensive evaluation of long-term validity 
. However, this approach is also questioned, given chronological changes in CVD event rates and associated risk factors 
, which are less likely to affect validity if more recent and thus shorter follow-up data are used 
. We did not evaluate the model's performance on predicting outcomes at the individual level (discrimination) and group level (calibration) using external data. This would be necessary to investigate to what extent the personalized predictions are transportable to other settings and geographical sites, but is beyond the scope of this study.
A second limitation is that the relative risk reducing effect of statin therapy was kept constant over age and various risk factor levels. Although a number of observational studies 
found that the protective effect of cholesterol lowering on CVD events decreases in individuals aged 70 to 89 y, this was not confirmed by experimental research 
. Meta-analyses of statin trials demonstrate that effects on cardiovascular events are fairly independent of various risk factor levels 
. These trials, however, predominantly included participants with elevated risk factor levels. In the Rotterdam Study, individuals with normal risk factor levels were also included, and it is not known whether the relative risk reduction will be different for these individuals with normal levels. Thus, we cannot exclude the possibility of a small overestimation of the statin therapy effect in those with normal risk levels.
A third limitation is that, although we did account for baseline statin use, we did not take into account initiation of statin therapy during follow-up. Omitting this information could lead to an underestimation of the effect of statin therapy. However, in the 1990s, mass screening for dyslipidemia was not advocated, and statins were prescribed only to patients with a history of CVD or with persistent severe dyslipidemia after dietary intervention 
. Follow-up examinations of the Rotterdam Study population in 1997 revealed that statin use was quite limited 
. Thus, underestimation of statin effects due to treatment drop-ins will be small.
A fourth limitation is that the RISC model's outcomes did not perfectly match with all the outcomes as evaluated within statin trials. Therefore, we were not able to model a statin effect on total stroke events, and solely modeled an effect on first ischemic and unspecified stroke. However, these stroke subtypes contribute to 92% of all first stroke events in the Rotterdam Study 
. In addition, we did not model a direct statin effect on CVD mortality by causes other than myocardial infarction and stroke. Although a reduction in sudden cardiac death—a major component of CVD mortality—is observed in symptomatic patients treated with statins, the effect for participants without manifest CVD seems negligible 
. Nevertheless, we cannot exclude a small underestimation of benefits due to these modeling choices.
A final limitation is that our results may not be generalizable to other populations. The RISC model's output on cardiovascular mortality was most compatible with a population resembling inhabitants of a low CVD risk region. This finding confirms results from another cohort study 
, suggesting that cardiovascular mortality in Dutch individuals is most similar to predictions from the low risk region SCORE equation. In addition, the generalizability of our results also depends on the competing rate of mortality due to other diseases. Our estimations of remaining life expectancy for females and males at the ages of 60, 65, and 80 y, however, reasonably match estimations for low CVD risk countries as projected by the Organisation for Economic Co-operation and Development 
. Thus, the web-based calculator and color charts should be used with caution in individuals from regions with higher CVD risk.
The competing mortality risks from other CVD and non-CVD causes of death, which were not affected by statin therapy, sometimes resulted in counterintuitive lifetime outcomes. For example, age is the most important factor increasing both the yearly probability of CHD and stroke events, and the fatality of these events. Thus, age is expected to increase the health benefit of statin therapy. However, in the Rotterdam Study, age is even more strongly associated with an increase in yearly mortality from other causes of death 
. Subsequently, changes with statin therapy in lifetime outcomes were smaller with increasing age, because prevented CHD and stroke events were also increasingly substituted by other fatal events.
Although the average gain in total life expectancy with statin therapy may seem small, it is larger than that calculated for some other preventive interventions targeted at the general population 
. One should recognize that gains were much larger in particular participants, and were averaged out by participants who never experienced CVD. It should also be acknowledged that with the benefits of statin therapy, the costs, side effects, and disutility of daily pill use are likely to be acceptable across various age groups and risk levels, especially in a “low statin cost era” 
. In addition, we observed that gains in CHD/stroke-free life expectancy were generally larger than those in total life expectancy. Two phenomena can explain this observation. First, a large proportion of the CHD and stroke events were not fatal. Gains in CHD/stroke-free life expectancy are mainly driven by statin effects on non-fatal CHD and stroke event rates, while gains in total life expectancy are driven by effects on CHD and stroke death rates. Second, individuals in whom fatal CHD and stroke events are avoided are also likely to be at elevated risk for death by other causes. Our finding of a smaller effect of statin therapy on total life expectancy than on CHD/stroke-free life expectancy is in agreement with the results from statin trials, in which generally only modest effects are demonstrated for crude total mortality risks, while effects on crude CHD and stroke incidence risks are more pronounced 
Currently, statin therapy choices are based on short-term CVD risk without statin therapy and the expected risk reduction with statin therapy over the same time period. We converted survival benefits with statin therapy into total life expectancy and CHD/stroke-free life expectancy.
We believe that the prediction of statin therapy effects on (disease-free) life expectancy can be complementary to the 10-y CVD risk assessment in two ways. First, instead of regarding a fixed time point, i.e., 10 y, the benefit of statin therapy considering the entire survival curve can be communicated by primary care physicians. Second, the benefit of statin therapy is calculated taking into account competing mortality risks. The potential value of personalizing the gain in total and CHD/stroke-free life expectancy with statin therapy is best illustrated by and . A substantial number of individuals with 10-y total CVD mortality risk lower than 5%, for whom statin therapy is generally not recommended according to current European Society of Cardiology guidelines, may benefit to the same extent as individuals with a high risk. A similar pattern will apply to predictions based on other CVD risk models, such as risk scores based on the Framingham Study 
, because these use the same risk factors, with effects pointing in equal directions.
When making decisions on statin therapy, the fact that the benefit in life expectancy diminishes with advancing age may be considered by physicians, especially in the elderly. If independently validated, physicians could use the web-based calculator and color charts to frame survival outcomes in different ways and to discuss them with the patient in light of the expected duration of statin use. The longer the life expectancy, and therefore the expected duration of statin use, the higher the costs and possibility of adverse effects. Besides the costs averted by CVD prevention, these important outcomes could influence the decision, but were not taken into account in our analysis. In addition, it should be acknowledged that the calculated differences in the personalized lifetime outcomes may vary across different clinical settings and are subject to the parameter uncertainty in the underlying decision model. These caveats would need to be discussed with patients when they are informed about the benefits of statin therapy.
In conclusion, we demonstrated that life expectancy benefits with statin therapy can be predicted using an individual's risk factor profile. The predicted gains in life expectancy we found are generally small. If the underlying model is validated in an independent cohort, the developed tools may be useful in discussing with patients their individual expected outcomes with statin therapy. Ideally, communication of personalized outcomes will ultimately result in better clinical outcomes. Improved understanding of potential gains will, however, not necessarily go hand in hand with an improvement in clinical outcomes, because patients could be less likely to choose statin therapy when more information on benefits is provided 
. In addition to an external validation of our predictions, personalized estimates for costs and side effects of statin therapy should be included in future research. Finally, the impact of communicating life expectancy benefits on satisfaction, behavioral, and clinical outcome measures should be studied.