Trial sequential analyses of six meta-analyses considering heterogeneity and trial weight

Tags: Poster
Thorlund K, Wetterslev J, Brok J, Gluud C

Background: Meta-analyses should not consider included patients as originating from one trial. Meta-analyses are rarely analyzed with trial sequential boundaries (TSB) to determine when strong evidence is reached (Pogue ControlledClinTrials 1997). TSB require calculation of optimal information size (OIS) based on trials with adequate quality to obtain appropriate information fraction (IF). Calculation of OIS does not consider heterogeneity among trials (I2) (Higgins StatMed 2002) and IF ignores the concept of pooled estimates in meta-analyses.

Objective: To examine meta-analyses for spurious p<0.05 values and spurious crossing of (TSB) with the Lan-DeMets discrete sequential boundaries considering heterogeneity and trial weight.

Methods: We randomly selected six meta-analyses with at least five trials reporting a binary primary outcome from the 171 systematic Cochrane Neonatal Review Group reviews in The Cochrane Library (Issue 2, 2004). We defined optimal heterogeneity-based information size (OHBIS=OIS/1-I2) based on I2 from all trials and weight fraction (WFi =weight[dot]n/OHBIS). The cumulated z-curve, standard TSB (TSBS), and heterogeneity-based TSB (TSBH) were constructed. TSB were considered for alpha=5% and 1-Beta=80% using both IF and WF.

Results: Three meta-analyses presented firm evidence as the cumulated z-curve crossed z=1.96. TSBS. and TSBH. Two z-curves crossed z=1.96 temporarily but returned to non-significant values. One of these also crossed TSBS temporarily, but never crossed TSBH. One z-curve never crossed the z=1.96.TSBS. or TSBH.

Conclusions: Three meta-analyses with p<0.05 were supported by z-curves crossing the TSBH. Spurious crossing of p<0.05 and TSBS may be eliminated by using TSBH on meta-analyses. TSBH compared to either p<0.05 or TSBS may reduce the risk of types I errors without increasing risk of type II errors.