Statins for Primary Prevention of Cardiovascular Disease

IN RESPONSE: Important comments were forwarded on our recently published article about risk thresholds to start a statin therapy for primary prevention of cardiovascular disease. Dr. Good and his colleagues, referring to the AFCAPS/TexCAPS study (1), suggested that the potency of statins may increase over time. We decided against considering such a time effect because we did not find consistent evidence of such a trend in our data and other systematic reviews (2). Dr. Spence agreed with our assumptions of increased risk for myopathy and diabetes due to statins but questioned the risks for cataract, hepatotoxicity, and hemorrhagic stroke. Dr. Zhou also raised similar arguments, particularly against the inclusion of observational studies. We are aware of the long-lasting debate on harms associated with statins and the wide range of estimates (from almost none to large effects). One may also argue for using postmarketing surveillance data on harms with larger effect estimates (3). All data sources would have their own advantages and limitations. This is why we used convergent estimates from high-quality trials and observational studies, which represent, in our view, the most valid and applicable estimates. Of note, trials contributed substantially more to convergent estimates because of their higher precision; therefore, the effect estimates we used for harms were rather small (see supplementary material in our article). In addition, we had determined outcomes a priori (4) and included them regardless of statistical significance (for example, the relative risk for cancer was close to 1.0 and was far from being statistically significant). Drs. Djulbegovic and Hozo raised 2 issuesthe appropriateness of the Gail approach to determine thresholds, and the accuracy of our thresholds (Table 2, second and third columns). They recalculated the thresholds using an alternative method (seventh column) (5). Reasons for the lower thresholds are that Drs. Djulbegovic and Hozo calculated the threshold at which benefit exactly equals harm and they considered competing risk only for harm outcomes, not for CVD, which overestimated the net benefit. We defined the threshold at the 10-year risk where benefit exceeds harm with a probability of at least 60%, because patients would opt to take statins only if the benefits exceed harms and not when they cancel each other out. To consider statistical uncertainty, we performed the analysis in a stochastic way by generating a distribution of benefitharm balance index and set the threshold where the benefits exceeded the harms in at least 60000 out of the 100000 benefitharm indices of that distribution. When considering competing risks for both benefits and harms as well as statistical uncertainty, Djulbegovic and Hozo's method showed similar values with our thresholds at 50% probability of net benefit (fourth and fifth columns). In our view, clinically actionable thresholds should ideally be defined where patients and physicians would take action, although this is not easy to define. We do not think that individuals would take statins if they were told that statins have a 50% probability of benefit exceeding harm; in other words, the probability of harm exceeding benefit is also 50%. Thus, we chose the 60% probability, although arbitrarily, to ensure that the benefits exceed the harms at least to some extent. Some individuals may choose even higher thresholds to ensure more benefits. Hence, a one-size-fits-all threshold is unlikely. With respect to this, Dr. Sorum is correct in his comment that the thresholds are preference sensitive (see supplementary materials in our article). This implies that individual preferences, rather than an average, should be considered, and decisions should be tailored to individuals that reflect their values and preferences. Table 2. 10-Year CVD Risk Thresholds Determined Using the Gail Approach and DjulbegovicHozo Exact Threshold Model