Thirteen studies (254,950 patients, range 67 to 121,254) were included in the review: one randomised trial (67 patients), 10 cohort studies (252,505 patients) and two case-control studies (2,378 patients). Nine non-randomised studies were assessed as being of high quality (scored 24 to 26) and three had intermediate quality (scored 21 to 22). The randomised trial was of unclear quality. Mortality data were reported at time intervals: in-hospital (four studies) and 30 days (six studies), 90 days (two studies) or six months follow-up (one study).
Overall mortality following pneumonia was statistically significantly lower among current statin users than with statin non-users (OR 0.62, 95% CI 0.54 to 0.71; 10 studies; Ι²=50%). Similar results were shown for each mortality time interval subgroup and the most pronounced effect was shown at 30 days follow-up (OR 0.42, 95% CI 0.25 to 0.71; four studies; Ι²=63.4%). No evidence of publication bias was found.
Adjustments were made for various factors, including social and demographic factors, pneumonia process measures, smoking and vaccination status, comorbidity indices and propensity to receive statin treatment. The adjusted pooled analysis demonstrated a statistically significantly lower risk of mortality among current statin users versus statin non-users (OR 0.66, 95% CI 0.55 to 0.79; 11 studies; Ι²=72.6%). Again, the most pronounced effect was shown at 30 days follow-up (OR 0.48, 95% CI 0.39 to 0.59; five studies; Ι²=0%). Numbers of deaths prevented per 1,000 statins users (range 3.4 to 70.3) were reported for various mortality risks (1% to 25%).
Most subgroup analyses demonstrated similar pooled results for the adjusted and unadjusted mortality outcomes. No effect of statin use on mortality was observed in the adjusted effect size for prospective cohort studies (three studies) or for the underpowered randomised trial. The strength of association between statin use and mortality substantially weakened when studies were analysed according to the inclusion of confounders within their models.
In sensitivity analyses, the pooled results for the adjusted and unadjusted mortality outcomes were not substantially altered.