【正文】 202 1 2 18 EORTC 08861 105 1 2 19 LILLE 163 1 2 20 GETCB 05CB86 539 3 2 Subtotal 1009 4 2 (Isquared = %, p = ) 4 2 4 Heterogeneity between groups: p = 5 Overall 1009 4 (Isquared = %, p = ) Estimates。 = ratio of tau178。 correlations) estimated simultaneously ? But basic analysis also possible with “twostage” ? Relative effect (interaction coefficient) only ? Same approach (inversevariance) as for main effects ? Ensures no estimation bias from betweentrial effects ? Can be presented in a forest plot, with assessment of heterogeneity etc. ? Discussed in a published paper (Fisher 2022) ?????? = ???? +???????? +???????? + ?????????????? “Onestage” vs “twostage” Onestage Twostage Pros All coeffs amp。 parison with metan ? Covariate interactions ? Combining AD with IPD ? Advanced syntax ? The forestplot mand ? Interface with ipdmetan ? Standalone use and “stacking” ? Summary and Conclusion Introduction to IPD metaanalysis ? Metaanalysis (MA): ? Use statistical methods to bine results of “similar” trials to give a single estimate of effect ? Increase power amp。 (%) | % Modified H178。 aggregate data as subgroups Trials included from IPD: 7 Patients included: 1333 Trials included from aggregate data: 4 Patients included: 1009 Pooling of main (treatment) effect estimate arm using Fixedeffects trial reference | number | Effect [95% Conf. Interval] % Weight + IPD | LCSG 773 | CAMS | ... | ... Subgroup effect | + Aggregate | belgium | EORTC 08861 | ... | ... Subgroup effect | Tests of effect size = 1: IPD z = p = Aggregate z = p = Inclusion of aggregate data: Screen output Inclusion of aggregate data: Forest plot I P DL C S G 7 7 3C A M SM R C L U 1 1S L O V E N I AG E TC B 0 4 C B 8 6I TA L YK O R E AS u b t o t a l ( I sq u a r e d = 0 . 0 % , p = 0 . 7 4 0 )A g g r e g a t eb e l g i u mE O R TC 0 8 8 6 1L I L L EG E TC B 0 5 C B 8 6S u b t o t a l ( I sq u a r e d = 0 . 0 % , p = 0 . 9 6 4 )r e f e r e n ce n u m b e rt r i a l1 . 1 2 ( 0 . 8 3 , 1 . 5 3 )1 . 0 3 ( 0 . 7 7 , 1 . 3 8 )0 . 9 6 ( 0 . 7 4 , 1 . 2 4 )0 . 8 9 ( 0 . 5 4 , 1 . 4 9 )1 . 1 4 ( 0 . 8 0 , 1 . 6 2 )0 . 6 9 ( 0 . 4 0 , 1 . 2 0 )1 . 1 6 ( 0 . 7 6 , 1 . 7 6 )1 . 0 2 ( 0 . 9 0 , 1 . 1 6 )1 . 4 6 ( 1 . 0 7 , 1 . 9 8 )1 . 6 4 ( 0 . 9 1 , 2 . 9 6 )1 . 5 7 ( 1 . 0 6 , 2 . 3 2 )1 . 4 4 ( 1 . 1 3 , 1 . 8 3 )1 . 4 8 ( 1 . 2 6 , 1 . 7 4 )E f f e ct ( 9 5 % C I )1 8 . 1 81 9 . 9 22 6 . 1 26 . 4 81 3 . 8 55 . 6 99 . 7 61 0 0 . 0 02 8 . 6 17 . 7 91 7 . 5 64 6 . 0 31 0 0 . 0 0W e i g h t%1 4. 2 5Advanced syntax example: non “eclass” estimation mand ipdmetan (u[1,1]/V[1,1]) (1/sqrt(V[1,1])) , study(trialid) eform ad(, byad) lcols(evrate=_d % Event rate) rcols(u[1,1] % oE(o) V[1,1] % V(o)) forest(nooverall nostats nowt) : sts test arm if subgroup==0, mat(u V) Effect estimate amp。 support ? Assorted colleagues for testing ? Reference: ? Fisher D. J. et al. 2022. Journal of Clinical Epidemiology 64: 94967