Perioperative statin therapy in cardiac surgery: a meta-analysis of randomized controlled trials

Two trained investigators (AP, AB) independently searched PubMed, the Cochrane Central Register of clinical trials, and EMBASE (last updated on 1 November 2016) for appropriate articles. The full PubMed search strategy is presented in the supplementary material (Additional file 1). The search strategy was designed to include any RCT ever performed with perioperative statin therapy compared to control in adult humans in a cardiac surgery setting. No language restriction was enforced. References for eligible studies and identified reviews were searched by hand.

Records obtained from searches were first independently examined at an abstract level by two trained investigators (AP and AB). Following the initial abstract assessment, all identified studies were acquired as full-text. Eligible studies met the following criteria defined as patient, population or problem, intervention, comparison, outcomes and study design (PICOS): (1) population: adult cardiac surgery patients; (2) intervention: administration of perioperative statin therapy; (3) comparison intervention: placebo or no active intervention as control; (4) outcome: any primary or secondary outcome of the present systematic review (see subsequent text); and (5) study design: randomized controlled trial. The exclusion criteria were pediatric studies and overlapping populations. Two investigators (AP and AB) independently assessed selected studies for the final analysis, with eventual divergences finally resolved by consensus with a third author (GL).

Two authors (AP and AB) independently extracted data from studies and entered them into a predefined database. Discrepancies were identified and resolved through discussion with a third author (GL) if necessary. We collected potential sources of significant clinical heterogeneity such as study design, clinical setting, details of the case and control interventions, data on the predefined outcomes, and information necessary to assess risk of bias.

The primary outcomes were postoperative AKI, postoperative AF, postoperative MI, postoperative stroke, and postoperative infection. The secondary outcome was mortality at the longest available follow up. The outcomes were reported in the present review as per-author definition. If the data on the postoperative outcomes were absent or incomplete, missing data were requested from the corresponding authors of the study. The data extraction followed the intention-to-treat basis whenever possible.

We used the Cochrane approach [23, 26] to evaluate the methodological quality of each included trial (Additional file 1). Each trial was finally judged to be of low, unclear, or high risk of bias. The quality of the evidence for each outcome was summarized with the grading of recommendations assessment, development, and evaluation (GRADE) method [23, 26, 27].

For each outcome, we calculated the odds ratio (OR) with 95% confidence intervals (CI). We reported the proportion of patients with the outcome in each group and the p value for the comparison between the groups. A p value <0.05 was considered significant. In the case of statistically significant ORs, we calculated the number needed to treat (NNT) or number needed to harm (NNH). The primary analysis of the present review was restricted to studies with low risk of bias, as suggested by the Cochrane Collaboration tool for assessing risk of bias [26].

Heterogeneity was explored by the Cochran Q statistic and characterized with I ². We used a fixed-effect model for meta-analysis in the absence of significant heterogeneity, defined as a p value >0.10 and I ² < 50%. In case of significant heterogeneity, we employed the random-effects model except if few trials dominated the available evidence or if significant publication bias was present, as random-effects meta-analysis in these contexts can give inappropriately high weight to smaller studies [23]. Two investigators (AP and AB) independently evaluated publication bias and small trials bias, analyzing a funnel plot and assessing the asymmetry in the funnel plot of trial size against treatment effect.

We performed sensitivity analyses for each outcome in order to assess the influence of risk of bias in the trials, including all eligible trials despite their risk of bias and including only trials with unclear or high risk of bias. In accordance with the Cochrane methodology [23], we performed sensitivity analysis for each outcome to investigate whether choice of summary statistic (OR, risk ratio (RR), risk difference (RD)) is critical to the results of the meta-analysis. We performed further sensitivity analysis for each outcome including only trials enrolling more than 200 patients or including only placebo-controlled studies.

Two authors independently evaluated the possibility of significant conflicts of interest within each study. They evaluated the funding of the study, the potential for authors’ conflicts of interests, the methodological quality of the study, and the positive/negative/indifferent findings of the study over statin. In case of possible or unclear industrial conflicts of interest among studies included in the analysis, we performed sensitivity analysis excluding them. The results of sensitivity analyses are reported only if significantly different from the primary analysis.

Post-hoc meta-regression was employed to examine the possible influence of length of preoperative therapy, proportion of CABG patients, trial size, and publication year on clinical outcomes in all eligible trials. Post-hoc subgroup analyses were performed on trials that included only statin-naïve patients, on trials enrolling mixed populations (statin-naïve and chronic statin therapy), and on trials that randomized patients on a postoperative statin regimen or not. Subgroup differences were tested using chi-square statistics [23]. The meta-analysis was performed using Review Manager (RevMan (Computer program), Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

Finally, to confirm the validity of our findings, we performed post-hoc trial sequential analysis (TSA) [28‐30], with the intent of maintaining an overall 5% risk of type I error and a 20% risk of type II error, at a power of 80%. Relative risk reduction (RRR) or relative risk increase (RRI) for each outcome was derived from the literature in order to evidence a clinically meaningful difference (Additional file 1). We used the Copenhagen Trial Unit TSA software (version 1.0, http://​www.​ctu.​dk/​tsa).

In total, 3699 references were examined. Major exclusions are presented in the supplementary data together with the reasons for exclusion (Additional file 1). Finally, 23 articles (5102 randomized patients) [21, 22, 31‐51] were included in the analysis (Fig. 1).

Characteristics of the 23 trials are listed in Table 1 and Additional file 1. Eligible trials included from 30 to 1922 patients and were all single-center studies. In six cases we received data from the authors on further outcomes [21, 31, 38, 40, 42, 51].

All trials included elective cardiac surgery patients and the most represented procedure was CABG, performed in 78.01% of the patients.

Statins were administered preoperatively in all trials, with the length of treatment ranging from 1 to 28 days (median 7 days). The total duration of statin treatment varied from 2 to 33 days; the variety of statin doses and regimens are shown in Table 1. The randomized treatment was also administered postoperatively in 8 trials [22, 34, 35, 38, 42, 44, 45, 51]. Atorvastatin (20 to 80 mg) was administered in 14 trials, Simvastatin 20 mg in 3 trials, Rosuvastatin 20 mg in 2 trials, Fluvastatin 80 mg in 1 trial, and Pravastatin 40 mg in 1 trial. In one case the statin regimen was not specified [41]. Placebo was administered as control in 16 trials and in 7 trials there was no intervention administered as control [31, 34, 37, 45, 49‐51].

Ten trials reported postoperative data on AKI (3354 patients), 19 trials on AF (4737 patients), 19 trials on MI (4283 patients), 12 trials on stroke (3631 patients), 7 trials on infections (2961 patients), and 18 trials on mortality (4157 patients).

Three trials were judged at low risk of bias in all bias domains [21, 22, 44]. Five trials were scored as having unclear risk of bias [35, 36, 38, 42, 46] and 15 trials were at high risk of bias (Fig. 2 and Additional file 1). The overall quality of evidence according to GRADE was reported for each outcome (Table 2 and Additional file 1).

When including trials with low risk of bias, the administration of perioperative statins was associated with increased incidence of postoperative AKI as compared with placebo (314 of 1318 patients (23.82%) in the statin group versus 262 of 1319 patients (19.86%) in the placebo group; OR 1.26 (95% CI 1.05–1.52); p = 0.01; NNH 25) (Fig. 3). The results are supported by the TSA, which showed firm evidence for a 25% RRI (Additional file 1). The overall quality of evidence was high according to GRADE.

The quantitative results of the systematic review are displayed below and in Table 2 and supplementary material (Additional file 1).

When including all the eligible trials despite the risk of bias, statin therapy was associated with a non-significant difference in AKI versus control (OR 1.18 (95% CI 0.99–1.41]; p = 0.06) (Fig. 3). Nonetheless, the latter analysis was characterized by possible small-study publication bias and on sensitivity analyses the incidence of AKI was higher with statin therapy (Additional file 1).

There was no difference in the rate of postoperative AF in trials with low risk of bias (318 of 1268 (25.07%) in the statin group and 300 of 1269 (23.64%) in the placebo group, OR 1.08 (95% CI 0.90–1.30); p = 0.40) (Fig. 4) and the TSA showed futility of the statin treatment when assuming an RRR of 20% (Additional file 1). The overall quality of evidence was moderate.

On the contrary, the analysis that also included trials with high and unclear risk of bias showed a lower incidence of AF among patients allocated to statins (Fig. 4), but TSA did not confirm the findings. There was significantly high heterogeneity (p for heterogeneity = 0.0004, I ² 60%) and important small-study publication bias; on sensitivity analyses there was no difference in AF between statins and control when including all eligible trials despite their risk of bias (Additional file 1).

The rate of postoperative MI did not change significantly between groups among trials with low risk of bias (86 of 1268 (6.78%) in the statin group versus 88 of 1269 (6.93%) in the placebo group; OR 0.90 (95% CI 0.57–1.42)) (Fig. 5) and TSA showed futility of the statin treatment when assuming an RRR of 30% (Additional file 1). The overall quality of evidence was moderate.

When including all trials regardless of the risk of bias, statin therapy was not associated with a difference in postoperative MI versus control (Fig. 5); publication bias was present (Additional file 1).

There was no difference in the rate of postoperative stroke in trials with low risk of bias (15 of 1268 (1.18%) in the statin group versus 12 of 1269 (0.95%) in the placebo group; OR 1.25 (95% CI, 0.58–2.70)) and TSA showed futility of the statin treatment when assuming an RRR of 80%. The overall quality of evidence was moderate. Including all trials, statin therapy was not associated with a difference in the rate of postoperative stroke (Additional file 1).

There was no difference in the rate of postoperative infections between the statin and placebo groups in trials with low risk of bias (88 of 1268 (6.94%) in the statin group versus 108 of 1269 (8.51%) in the placebo group; OR 0.80 (95% CI 0.60, 1.07)); on meta-analysis including also trials with higher risk of bias there was no significant difference between groups (Additional file 1). TSA showed futility of the statin treatment only when including all the eligible trials despite their risk of bias (Additional file 1). The overall quality of evidence was low.

The administration of perioperative statins was not associated with a significant difference in hospital mortality in trials with low risk of bias (9 of 1318 (0.68%) in the statin group versus 2 of 1319 (0.15%) in the placebo group; OR 1.26 [95% CI, 1.05-1.52]; p = 0.06) (Fig. 6); TSA did not support this as firm evidence. The overall quality of evidence was low. The sensitivity analysis showed a small increase in hospital mortality in the statin group when changing the summary statistic (RD 0.01 (95%CI 0.00–0.01); p 0.04) (Additional file 1). There was no difference in short-term mortality when including all eligible trials despite their risk of bias (Fig. 6).

Meta-regression analysis of trials with low risk of bias did not identify possible relationships between the length of the preoperative regimen and clinical outcomes. Meta-regression including all trials revealed a possible relationship between the length of the preoperative regimen and AF, with patients on longer preoperative regimens having a lower incidence of AF then patients on shorter regimens (p = 0.024). However, the latter analysis was driven mainly by the results of trials with unclear and high risk of bias (Additional file 1).

Subgroup analysis including all trials addressing the effect of the presence or lack of a postoperative regimen identified a significant subgroup difference for AKI (chi-squared 4.68, p = 0.03) and AF (chi-squared 19.21, p < 0.0001), suggesting a better outcome in patients not taking postoperative statins. However, the analysis has important limitations, as few trials did not administer postoperative therapy and all these trials had high or unclear risk of bias (Additional file 1).

Among trials with low risk of bias, the only available outcome to estimate was AF (2 trials, 2537 patients) with no significant difference between statin-naïve patients or those on chronic therapy (chi-squared 0.05, p = 0.82). When including all the trials (19 trials, 4218 patients), there were no significant differences between trials including statin-naïve or mixed populations, except for AF (chi-squared 5.69, p = 0.02). However, this analysis should be interpreted with caution, because all the trials enrolling only statin-naïve patients had high or unclear risk of bias (Additional file 1).

Meta-regression analysis did not reveal significant correlation between the proportion of patients undergoing CABG and the log-OR of any postoperative clinical outcomes analyzed, even when including trials with higher risk of bias (Additional file 1).

Meta-regression including all trials demonstrated the influence of trial size (p = 0.02) and publication year (p < 0.01) on postoperative AF and AKI (Additional file 1). This underlines that: (1) AKI and AF rates changed over years, with results from older studies in favor of statins; and (2) AKI and AF rates changed in relation to trial sample size, with results from smaller trials in favor of statins.

The presence of possible publication bias was suggested when assessing the asymmetry of the funnel plot including all the eligible trials; in particular, publication bias was evident in AKI, AF, and MI (see previous), suggesting that small studies tended to report larger treatment effects in favor of statin therapy than larger studies did.

Meta-analysis including trials with high and unclear risk of bias (n = 20, 2465 patients) showed lower incidence of AF in the statin group and no significant difference in other postoperative outcomes (Additional file 1).

One of the preferable meta-analytical strategies is to restrict the primary analysis to studies at low risk of bias [23, 26]. The choice should be based on the balance between the potential for bias and the loss of precision when studies at high and unclear risk of bias are excluded [26]. Among the randomized literature, we found significant small-study publication bias and serious differences in risk of bias within studies. We must recognize that publication bias is common in peer-reviewed journals and particularly in critical care medicine [70], because positive studies are easier and more attractive to publish than neutral or negative studies [71], and small trials are more likely to report larger beneficial effects than large trials, which could be partly explained by the lower methodological quality in smaller trials [70]. It is noteworthy that the largest trials with low risk of bias performed so far on the topic had neutral or negative results from analysis of the use of perioperative statins in cardiac surgery patients [21, 22]. To this end, we think that our per-protocol primary analysis, including only randomized placebo-controlled trials with low risk of bias, could have limited this problem. However, only three trials with low risk of bias with a total of 2637 patients have been published so far and included in our primary analysis.

There is some degree of variability in the nature of statin therapy, because statins may be administered using different strategies, some of which may be more effective than others; for example, there is non-randomized evidence to support an increased rate of AKI among patients taking high-potency statins, with the strongest rate in the first 4 months after initiation of treatment [72]. We did not perform subgroup analysis of different types of perioperative statin regimens, because almost all the trials administered different statin doses and formulations for different lengths of time, and the analysis would have been biased by results of trials with higher risk of bias. Finally, our mortality analysis included short-term mortality (in-hospital and 30-day mortality), addressing the fact that mortality should be assessed after longer follow up, in order to evidence the long-term effects of the interventions.

This systematic review synthesized evidence from RCTs and may help the execution of future clinical studies assessing the exact time point for interruption of preoperative statin therapy in patients already taking statins. Further RCTs should systematically evaluate the relationship between postoperative outcomes and variables related to the patient (e.g., chronic kidney disease, statin-naïvety), to the cardiac disease (e.g., coronary artery disease), to the surgical procedure (e.g., off-pump versus on-pump surgery), and to the specific statin regimen.