Systematic Review and Meta-Analysis on the Association between Outpatient Statins Use and Infectious Disease-Related Mortality

Background To update and refine systematic literature review on the association between outpatient statins use and mortality in patients with infectious disease. Materials and Methods We searched articles published before September 31, 2012, on the association between statins and infectious disease-related mortality through electronic databases. Eligible articles were analyzed in Review Manager 5.1. We conducted stratification analysis by study design, infection types, clinical outcomes and study locations. Results The pooled odds ratio (OR) for death (statins use vs. no use) across the 41 included studies was 0.71 (95% confidence interval: 0.64, 0.78). The corresponding pooled ORs were 0.58 (0.38, 0.90), 0.66 (0.57, 0.75), 0.71 (0.57, 0.89) and 0.83 (0.67, 1.04) for the case-control study, retrospective cohort studies, prospective cohort studies and RCTs; 0.40 (0.20, 0.78), 0.61 (0.41, 0.90), 0.69 (0.62, 0.78) and 0.86 (0.68, 1.09) for bacteremia, sepsis, pneumonia and other infections; 0.62 (0.534, 0.72), 0.68 (0.53, 0.89), 0.71 (0.61, 0.83) and 0.86 (0.70, 1.07) for 30-day, 90-day, in-hospital and long-term (>1 year) mortality, respectively. Conclusions Outpatient statins use is associated with a lower risk of death in patients with infectious disease in observational studies, but in a less extent in clinical trials. This association also varies considerably by infection types and clinical outcomes.


Introduction
Severe infections always remain a major cause of morbidity, mortality, and economic burden worldwide. For example, there are nearly 751,000 cases of sepsis in the United States each year and the number of cases is still increasing by 1.5% per year, resulting in an annual estimated cost of $16.7 billions [1]. Despite new advances in antimicrobial therapy and medical management, only 50%-70% patients can survive from sepsis [2]. In addition, Pneumonia induced by influenza and chronic obstructive pulmonary disease (COPD) had caused thousands of deaths in several epidemics [3], such as 1918 influenza epidemic, Asia influenza during 1957-1958, Hong Kong influenza in 1968, and Spanish influenza in 1942.
Statins, as one of the inhibitors of 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMG-CoA), are traditionally used to lower the level of blood cholesterol in patients with cardiovascular diseases or to prevent cardiovascular events. Recently, statins have been proposed as novel therapeutic and preventive agents for infection, given the increasing evidence, mostly from observational studies, that statins are associated with a lower mortality in patients with infectious disease [4][5][6][7]. Researchers believe that statins can mitigate the inflammatory response in patients with sepsis or COPD, which reflects its anti-inflammatory and immunoregulation effects [8]. This may occur through several possible biological mechanisms. Firstly, statins block the mevalonate pathway by inhibiting HMG-CoA reductase, which interfere with the recognition of microbial products by immune cell. They can also decrease production of proinflammatory cytokines such as tumor necrosis factor a (TNF-a), interleukin1 (IL-1), and IL-6 present during sepsis and COPD, and thus depress the inflammatory cascade [9]. Secondly, the antioxidant and anti-apoptotic properties of statins blunt the effects of sepsis [10]. Thirdly, the antithrombotic properties of statins decrease the effect of sepsisinduced coagulopathy [11]. Finally, statins increase the physiologic concentrations of nitric oxide (NO) by increasing the expression of endothelial NO synthase and down-regulating inducible NO synthase, and thus reverse the endothelial dysfunction in sepsis [12].These pleiotropic effects of statins have been demonstrated in experimental models (in vitro and in vivo), and some [13][14][15] but not all studies [4,6,7,16,17] showed serendipitous benefits of statins to patients with severe infections, such as sepsis and COPD. However, it is unclear which of them can explain the association between statins and lower infectious disease-related mortality observed in clinical observational studies.
So far, there are 4 published systematic reviews (3 of them are quantitative analyses) on the studies on the association between statins and infectious disease-related mortality [18][19][20][21]. Tleyjeh et al. [18] reviewed 9 cohort studies published as of 2007 that examined the effect of statins on infection-related mortality by bacteremia (n = 3), pneumonia (n = 3), sepsis (n = 2), and bacterial infection (n = 1). The pooled effect estimate (odds ratio or hazard ratio of mortality) was 0.55 (95% CI, 0.36-0.83) in favor of statins. Kopterides et al. [19] reviewed 15 studies published as of 2008 that examined statins and infection-related mortality: 10 of them reported protective effects of statins, 4 showed null effects, and 1 showed risk/adverse effects. Janda et al. [20]  . However, these systematic reviews have not sufficiently addressed the 3 critical challenges in this field: potential time-varying effects of statins, the differential effects of statins by types of infections, conflicting by results in different study design. Answers to these 3 questions have important implications: they can help us to better interpret (e.g. causality) on the observed associations, and can also influence how physicians, patients, and public health policy makers use the existing evidence. Moreover, only 1 randomized clinical trial (RCT) was included in previous reviews. Given the high validity of RCTs, it is important to include more recently published RCTs into the review/meta-analysis to provide a more conclusive interpretation on the potential effects of statins on infectious disease-related mortality.
Therefore, this meta-analysis aimed 1) to update the literature as of September 2012; 2) to summarize the association between outpatient statins use and infectious disease-related mortality across all published studies (observational studies and RCTs); 3) to examine whether this association differ by study design, infection types, outcome measures and study locations.

Methods
The systematic review and meta-analysis was performed according to the recently published recommendations and checklist of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.

Search Strategy
All searches were limited to ''English language'' and ''humans''. Two researhers conducted literature search independently in 4 steps. Step  Step 2, we excluded reviews, letter, comments, editorials, corrected publications, duplicate publications and abstracts only.
Step 3, we screened all the remaining abstracts and selected those that met our eligibility criteria (see exclusion criteria below).
Step 4, we searched the full-text articles of these eligible abstracts.

Study Selection
Our included surveillance studies, cohort studies, case-control studies, and RCTs that focused on statins and infectious diseases. Infectious diseases in this Meta-analysis included phenomena, COPD, sepsis, bacteremia, respiratory infection, organ transplantation or any open surgery induced infection. Outpatient statins use was defined as use of statins before the admission or both before and during the hospitalization. We excluded studies that, 1) only focused on the preventive effects of statins on the incidence of infectious disease; 2) only had information on statins use during the hospitation for observational study; 3) did not provide clear definition of statins use; 4)had no outcome measures for infectionrelated mortality.

Data Extraction and Quality Assessment
Two researchers (Ma Y and Peng J) independently reviewed the included studies and used the same Excel spreadsheet to extract relevant information of each study, including study objective, sample demographics (i.e. age, gender), comorbidities, type of infection, timing and dose of statins use, use of other medications or treatments, outcome measures (30-day, 90-day, in hospital and long-term mortality), association scale (OR or HR), and adjusted confounders. Disagreement between the 2 reviewers was first resolved by consensus. If consensus could not be reached, a senior reviewer (Guo Z) was consulted for the final decision. For observational studies, we used Newcastle-Ottawa Quality Assessment Scale (NOS) [32] to assess their quality. NOS has 8 questions reflecting 3 domains: quality of subject selection, comparability between two groups, and reliability of exposure or clinical outcomes. The full score of NOS is 9, and a score 8-9 is rated as excellent, 6-7 as good, 5 or below as fair. For clinical trials, we used Jadad Score [33] with 3 questions to assess the quality of randomization, blinding, and withdrawals or dropouts The full Jadad Score is 5, with 3-5 being rated as high quality and 1 or 2 as low quality.

Statistical Analysis
We conducted quantitative data analyses in Review Manager 5.1 (StataCorp, College Station, Tex). We used I 2 statistic to test heterogeneity across the included studies. An I 2 value of 0% indicates absence of heterogeneity, while 25%, 50%, and 75% indicates low, moderate, and high heterogeneity, respectively [34]. Random effects model was used for high level of heterogeneity and fix effects model was used for low and moderate levels of heterogeneity. We estimated pooled odds ratio (OR) and the corresponding 95% confidence interval (CI) across studies by weighting the Odds ratios of each individual study according to their log-transformed inverse variance. We also conducted subgroup analyses by study design, infection types, outcome measures (30-day, 90-day, in-hospital mortality and long-term mortality), and study locations (grouped by continents). Whenever possible, we used the adjusted OR instead of the crude OR to calculate pooled ORs. We conducted sensitivity analysis with failsafe number (Nfs) to calculate the number of negative studies needed to get opposite association between statins and mortality in patient with infectious disease. We also excluded single study or groups of studies based on the study design, outcomes, and quality score; re-ran the analysis to get new pooled ORs; and then compared the new pooled ORs with the original OR for all studies. This comparison can help to evaluate the appropriateness  of inclusion and exclusion criteria as well as the stability of included studies. We assessed potential publication bias with Egger precision weighted linear regression tests and funnel plots [35].
According to NOS, the 30 included observational studies were rated as excellent or good quality (score range, 7-9; mean, 8.15) ( Table 2). According to Jadad scale, the 10 included clinical trials were rated as high quality (score range, 4-5; mean, 4.7), which represented good quality of these studies.
The OR for death (statins use vs. no use) reported in the included studies ranged from 0.06 to 1.50 (Table 2). Twenty three studies found a protective effect of statins against death from infection, but the other 18 studies found null effects. There was substantial heterogeneity across the 41 studies (I 2 value, 74%), which supported the use of random effect model for meta-analysis. The overall pooled OR was 0.71 (95% CI: 0.64, 0.78) (Figure 2). Table 3 (Table 3).
There was substantial degree of heterogeneity across the included 41 studies (I 2 statistic, 74%). The I 2 statistic did not change considerably in the subgroup analyses based on total quality score, or subcategories of the quality score, i.e. exposure definition, outcomes, and confounding assessment. But the I 2 statistic was as low as 41% and 55%for clinical trials and studies in Europe, respectively. In our sensitivity analysis, there was no significant change in pooled OR when excluding any of the studies (data not shown) or any group of studies by infection type. Another sensitivity analysis showed that 938 negative studies were needed for getting the opposite association between statins and mortality in patient with infectious disease (Nfs = 938) and tolerance level was 103. Our Egger precision weighted linear regression tests showed the existence of publication bias (P-value ,0.0001). Funnel plot also showed the absence of small studies in which statins might increase infectious disease-related mortality (Figure 3).

Discussion
In this meta-analysis, we systematically reviewed 41 studies published during 2001-2012 on the association between statins use and infectious disease-related mortality. Overall, most observational studies found that statins were associated with lower mortality from infectious disease. Our pooled OR among these observational studies was similar to those in 3 previous metaanalyses by Tleyheh et al. [18], Surinder Janda et al. [20] and Bjorkhem et al [21]. However, we did not find conclusive evidence on this beneficial effect in clinical trials. In subgroup analysis, statins use was associated with lower 30-day, 90-day, and inhospital mortality, but not with long-term mortality. Statins use was associated with lower mortality from bacteremia, pneumonia, and sepsis, but not with mortality from other infections and intensive care unit (ICU) patients.
We found that the magnitude of the association between statins use and infectious disease-related mortality tended to decrease with time, i.e. strongest for 30-day mortality followed by 90-day mortality and in-hospital mortality, and null for long-term mortality. This time trend suggests that the beneficial effect (if exists) of statins to lower infectious-disease related mortality may be short-term only. However, this time trend should be interpreted with caution due to potential misclassification of death outcome based on medical records, especially for in-hospital mortality. The substantial variation in hospitalization length posed a big challenge on estimating and interpreting pooled effect size of statins use, which can be confirmed by heterogeneity test within hospitalization period subgroup (I 2 is 78%, P,0.0001).
We also found the magnitude of the association between statins use and mortality was strongest in patients with bacteremia. This suggests that the protective effect of statins may be superior for bacteremia than other types of infection. Alternatively, this may be partially explained by less use of antibiotics in patients with bacteremia: only 1 out of 6 studies on bacteremia patients, reported antibiotics use before the admission to hospital or during hospital, compared with most studies among patients with other types of infections (such as 100% in Kwong study and 20% in Yende study).
In this analysis, we found the reported effects of statins use varied strikingly by study design. Observational studies consistently supported the beneficial effect of statins in lowering infection disease mortality, while all RCTs showed null effects. Though RCTs often provide more valid results, these 10 RCTs can not completely outweigh the evidence on the beneficial effect of statins from observational studies. These ''null-effect'' RCTs are often criticized for small sample size [58]. These criticisms usually are based on two factors. Firstly, the formal sample size calculations that compute the numbers of patients required prospectively, as if the trial had not yet been carried out. Secondly, the true power calculated when trial is over. Based on alpha level, sample size and actual rates of primary events among control and experimental patients, we did post-hoc analysis by re-calculating the power of 8 negative trials in 10 RCTs and found that only one trial's retrospective power is bigger than 80% (Serrugs 81.5%), and all other 7 ones had power smaller than 80% (Fellstrom 8.9%, Kjekshus 25.5%, Wanner 14.3%, Stegmayr 47.3%, Holdaas 70.9%, Makris 60.5%). This under-power might be caused by the over-estimated event rate in control group and/or risk reduction level. In these 7 trials, investigator assumed the statins may reduce the risk of primary outcome by 20% to 50%, but the truth is the observed related risk reduction is obviously lower than 20% except for Makris study. This result indicated that the evidence to make a negative conclusion of statins is not sufficient and suggested retrospective sample size calculation is needed so as to add more representative patients in the trial. The sample size recalculation also showed that we still need 3-10 times of current number in each RCT to get positive result. Beside that, almost all these RCTs enrolled the patients who had existing cardiovascular disease. This could lead to misclassification of infection-related death, because of the difference between the primary and secondary infection. In addition, infection-related mortality is not a primary outcome in these studies on cardiovascular outcomes, and thus might not be measured accurately or appropriately.
Unexpectedly, we did not find significantly protective effect of statins against infection-related death in the 9 studies that focused on severe patient or patients admitted to ICU [7,23,26,28,36,39,40,51,57]  One explanation can be that those severe patients usually have many severe complications [26,39]. These complications often lead to adverse outcome and significantly increase the risk of death, which may dim the moderate protective effects of statins. Most interested us is that the result from only one clinical trial in these 9 studies provided evidence supporting that statins might affect the course of critically ill patients and decrees the ICU mortality [57]. Considering the differences in study design and implement between the observational studies and clinical trials, the most probable reason for this opposite result is the opportune moment of using statins. Patients in the observational studies might have already been using statins to treat high cholesterol (prevalent users). In clinical trials, however, statins were randomized to two groups after patients being recruited (new users). Another reason is the time difference between the progresses of critically ill and the onset time of statins. The immunomodulatory effects of statins can occur within 24 hours and thus acute treatment may down-regulate the level of pro-inflammatory cytokines [59]. That is why the acute curative effect in clinical trial is better than long-term effect in observational studies. These findings suggest that long-term use of statins may not be able to protect severe ICU patients against death from infection as common or less severe patients. However, we need more evidence from RCTs to confirm the acute effect of statin in reducing the ICU mortality among ICU patient.
The moderate heterogeneity across the included 41 studies may come from 2 main sources: study population and methodology. The significantly lower I 2 statistics in Europe studies than that in studies from other continents revealed the substantial difference in study population (e.g. ethnics and study locations), especially for limited source from Asia and Oceania population. For methodology, the low I 2 statistics in clinical trails indicated better quality control and more reliable results than observational studies. Other methodological heterogeneity included type of patients, type of infection, and dose of statins use across different studies.

Study Strengths
This meta-analysis had several strengths. First, we included much more RCTs than previous reviews (8 vs Inclusion of these RCTs can substantially improve the validity of our analysis. Secondly, we conducted subgroup analyses by 4 important factors, i.e. study design, types of infections, outcome measures, and study locations. These subgroups analyses can help us to better assess the sources of variation or inconsistency of findings, and also better understand the specific subgroups of patients that may benefit more or less from statins. Thirdly, we assessed the quality of each study by well-established score scales (NOS and Jadad Score). The relatively high quality of most included studies can improve our interpretation of the pooled effect estimates for stains use.

Study Limitations
Several limitations of this meta-analysis needed to be mentioned. First, we only included electronic database and published articles. Both Egger's test and asymmetry of funnel plotpotential suggested the existence of publication bias. Second, this meta- Figure 2. Forest plot of the association between statins and mortality for patients with infectious disease, by types of infection. Note: Each comparison was presented by the name of the first author and the year of the publication. The studies were shown by a point estimate of the OR and the accompanying 95% CI which were displayed on a logarithmic scale using a random effects model. The studies are sorted according to the estimate of OR. Between-study heterogeneity was tested by the x 2 -based Q-statistic, and its impact was quantified by I 2 which can range between 0 and 100%. doi:10.1371/journal.pone.0051548.g002 analysis included more observational studies (n = 31) than clinical trials (n = 10). We did not assign different weighting to included studies based on the validity of their study design (RCT vs. observational studies). Third, 6 of the 10 included RCTs were designed to test the effect of statins on cardiac outcomes rather than infectious disease-related mortality. So the validity of estimated associations from these RCTs may be comprised. Fortunately, several ongoing clinical trials [60][61][62][63] (ID: NCT00528580, NCT00979121, NCT00702130, NCT00676897, http://www.clinicaltrials.gov) aim to specifically examine the potential clinical benefit of statins in sepsis. We expect these studies will yield more conclusive evidence on this important topic in near future.

Conclusion
Based on this meta-analysis, we conclude that statins are associated with a lower risk of death in patients with infectious diseases in observational studies, but less in clinical trials. This beneficial effect tends to be short-term only. It seems to be stronger in patients with bacteremia but less for ICU patients with severe infection. More worldwide clinical trials specifically on this topic are urgently needed to provide more conclusive guideline for clinical practice.

Supporting Information
Checklist S1 PRISMA Checklist.