Evidence To Support Continuation of Statin Therapy in Patients with Staphylococcus aureus Bacteremia

DISCUSSION

Recent statin initiation with continuation of statin therapy for at least 3 days after culture was associated with a substantial protective effect on mortality among our large, national, real-world cohort with S. aureus bacteremia. These findings were robust in our primary analyses using propensity score matching and in our sensitivity analyses using propensity score quintile adjustment and inverse probability of treatment weighting. In vitro research suggests statins confer protective effects in S. aureus bacteremia, since they (i) inhibit S. aureus invasion of human endothelial cells (2, 11); (ii) interfere with S. aureus biofilm formation (16); and (iii) enhance clearance of S. aureus by phagocytes through the induction of DNA-based extracellular traps (12). Consistent with our findings, several meta-analyses have identified protective effects with statins on all-cause mortality among patients with various types of infections. Pleiotropic effects with statins were evaluated among patients with sepsis, pneumonia, or bacteremia by pooling 20 published studies (13). The authors reported a 50% reduced mortality in statin users (pooled odds ratio [OR], 0.49; 95% CI, 0.37 to 0.61). The bacteremia-related mortality (evaluated in 4 studies out of 20) was also significantly lower in statin users (pooled OR, 0.33; 95% CI, 0.09 to 0.75). Another meta-analysis found that outpatient use of statins was associated with a 29% decreased risk of all-cause mortality in patients with any infection (pooled OR across 41 studies, 0.71; 95% CI, 0.64 to 0.78) (14).

Among the included studies in both meta-analyses, exposure periods prior to hospitalization (pretreated) and after hospitalization (continuation and de novo) varied widely, and sensitivity analyses by statin exposure timing and duration were not conducted (13, 14). Indeed, some studies have included patients with such varied statin exposures, and application of the study findings to clinical practice would not be possible. One observational study defined statin use as the presence of a statin on the day of culture regardless of previous or continued use (17). This statin exposure definition combined both prevalent (of unknown timing and duration) and incident statin users, as well as patients continuing and not continuing statin therapy. Not surprisingly, statin use in this study was not associated with reductions in 90-day mortality, intensive care unit (ICU) admission, or hospital/ICU discharge when adjusting for confounders, including indications for statin therapy, using propensity score methods (17).

In our study, pretreated patients who continued on statin therapy experienced decreased rates of mortality, while these protective effects were not observed in pretreated patients who did not continue statin therapy or in patients with de novo use. These results support statin continuation through the period of inflammation, as effects on the inflammatory response are no longer observed once the statin is discontinued (18). Similar results were observed in a multicenter randomized placebo-controlled trial of 250 patients with severe sepsis assigned to statin therapy (n = 123) or placebo (n = 127) (8). Randomization accounted for prior statin use, defined as at least 2 weeks of statin use prior to hospitalization (prevalent users) or no use in the 2 weeks before admission; those with less than 2 weeks of statin use prior to admission were excluded. Pretreated statin users assigned to statin therapy had a lower 28-day mortality (5% versus 11%; P = 0.01) than the placebo group, although like our study, inpatient mortality was not significantly lower. Further, 28-day mortality in de novo users was similar to that of the placebo group (16.3% versus 14.9%; P = 0.78). It should be noted that duration of previous statin use was not assessed in the clinical trial, and as such, variations in outcomes may have existed because of duration, although the study size was likely too small to detect any such differences (pretreated assigned to statins, n = 37; pretreated assigned to placebo, n = 40).

We only know of one other study specifically examining the effects of statins on patient mortality in S. aureus bacteremia (19). A prospective cohort study, which included 160 S. aureus bacteremia episodes from one hospital in Spain, found that the 33 statin users were less likely to die within 14 days than nonusers (adjusted OR, 0.08; 95% CI, 0.01 to 0.66), but a significant difference between groups was not observed for 30-day mortality (adjusted OR, 0.35; 95% CI, 0.10 to 1.23; P = 0.10). Statin exposure was defined as prevalent statin use at bacteremia onset, and all users had at least 1 month of previous statin therapy. Another limitation of this Spanish study, besides prevalent statin use, was that 23/33 (70%) of the statin users had a vascular catheter as the source of bacteremia, compared to only 46/127 (36%) in nonusers. Given that vascular catheters are a readily removable source of bacteremia with lower mortality rates than other sources, such a difference is difficult to ignore (20). In our study, catheter source was similar between statin exposure groups and nonusers (Table 1).

Although most observational studies have confirmed the protective effects of statins on clinical outcomes in bacterial infections (19, 21–23), there is a concern surrounding this association due to the possibility of healthy user bias (24, 25). Patients taking preventive medications, such as statins, are more likely to have healthier behaviors resulting in favorable outcomes, including lower mortality rates, than sicker patients (26, 27). A multicenter inception cohort study conducted by Yende et al. supported this trend among statin users, providing evidence that statin use was significantly associated with good health behaviors, including health insurance, good functional status, and immunizations (28). Our approach to minimizing healthy user bias in our study was 3-fold (29). First, we designed our study to include only incident statin users and to assess patients continuing statin therapy as one exposure group and those not continuing as a separate exposure group, both of which were compared to a common reference group of nonusers. Second, we included proxies for healthy behaviors in our propensity score model, including use of preventative services (e.g., vaccination and health screenings) and conditions that impact health behaviors. Third, we implemented propensity score matching to identify nonusers with similar distributions of important patient characteristics related to health. By excluding prevalent statin users, we believe our study minimized the potential for healthy user bias, as this bias is observed in chronic medication use (25).

There are limitations to our study. First, although we employed propensity score methods to address potential confounders of the association between the use of statins and clinical outcomes, we were unable to control for unmeasured confounding. These methods allowed us to balance confounders of the exposure-outcome relationship that were included in the propensity score; however, it could not control for unbalanced factors that were not measured in our study. Second, variations in point estimates were observed with propensity score matching, adjustment, and inverse probability of treatment weighting. Although propensity score matching produced the most conservative estimates, it also resulted in the greatest balance between groups. Third, we attempted to identify incident statin use in order to assess the effect of statins at the time of S. aureus infection. We defined incident use as initiation in the 30 days prior to culture, with no prior statin exposure in the previous year. As such, incident use did not necessarily mean throughout the patient's lifetime. Therefore, our estimates may not completely rule out the influence of historical statin use (beyond the window that we defined in this study) on the outcomes. Fourth, our study results should be applied carefully to the general population, since our study was conducted among veterans and approximately 98% were male. Fifth, as a retrospective study of existing data, the accuracy of operational definitions depends on the data source. Although we utilized one of the most comprehensive and accurate data sources for health outcome research available in the United States, misclassification may still occur. For example, culture source is a free text field in the microbiology data, and without mention of a catheter in that field we could not determine whether it was a catheter source. Lastly, we did not assess outcomes for specific statins or doses, which is an important area of inquiry, as some data suggest added benefit of high-potency or high-dose statins (30, 31).