The efficacy of activated protein C for the treatment of sepsis: incorporating observational evidence with a Bayesian approach

Abstract

Objective: The present study aimed to combine observational evidence with randomised controlled trials (RCTs) by using the Bayesian approach.

Data sources: Electronic databases, including PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), ISI Web of Science, EMBASE and EBSCO were searched from inception to January 2014.

Study eligibility: RCTs and observational studies (OS) investigating the effectiveness of activated protein C (aPC) on mortality reduction were included for analysis.

Participants: Patients with sepsis.

Intervention: aPC.

Synthesis methods: Observational evidence was incorporated into the analysis by using power transformed priors in a Bayesian. Trial sequential analysis was performed to examine changes over time and whether further studies need to be conducted.

Main results: a total of 7 RCTs and 12 OS were included for the analysis. There was moderate heterogeneity among included RCTs (I(2)=48.6%, p=0.07). The pooled OR for mortality from RCTs was 1.00 (95% CI 0.84 to 1.19). In OS, there was potential publication bias as indicated by the funnel plot and the pooled OR for mortality with the use of aPC was 0.67 (95% CI 0.62 to 0.72). The pooled effect sizes of RCTs were changed by using different power transform priors derived from observational evidence. When observational evidence was used at its 'face value', the treatment effect of aPC was statistically significant in reducing mortality.

Conclusions: while RCT evidence showed no beneficial effect of aPC on sepsis, observational evidence showed a significant treatment effect of aPC. By using power transform priors in Bayesian model, we explicitly demonstrated how RCT evidence could be changed by observational evidence.

Trial registration number: The protocol for the current study was registered in PROSPERO (registration number: CRD42014009562).

Figure 1

Flow chart of study selection (RCTs, randomised controlled trials).

Figure 2

Forest plots showing the efficacy of activated protein C (aPC) on mortality reduction, reported separately for randomised controlled trials and observational studies.

Figure 3

Contour funnel plots showing the publication bias in randomised controlled trials (RCTs) and observational studies. Publication bias was identified for observational studies as reflected by the asymmetrically distributed component studies.

Figure 4

Sequential trial analysis involving randomised controlled trials showing that the Z-score crossed the futility line after the study Dhainaut et al. Parameters used for the creation of boundaries were—type: two-sided; type 1 error: 5%; α spending: O'Brien-Fleming; information axis: sample size; power: 80%; effect type intervention: relative risk reduction User Defined (15%). The shaded area indicates futility area. aPC, activated protein C.

Figure 5

Mean OR and 95% credible interval for different power transformation priors to down-weight observational evidence on the risk of death with aPC. Lower values of α down-weight the observational evidence.

Figure 6

Prior distribution derived by discounting observational evidence with α from 0.000001 to 1. The plots show that the precision of prior increases with increasing α values.