L11 Rev Ev Flashcards
LO2: def and desc sys Rev. Purp of sys Rev and circumstances when done.
-clearly focused Q. Explicit statements about types study, pp’s, intervenes and outc meas.
Sys lit search- clear meth, can criticise. Selec of mater. Appraisal. Synth (poss incl meta anal). So is v credit source ev-explicit, repro, transpar. Sys Rev= overview of prim studies that used explicit repro meth.
LO3: def and desc meta anal.
-meta anal= quant synth of res of 2+ prim studies that test same hyp in same way.
Math tech combs res= overall meas of tx effs.
-purp- facil synth lot studies, clear selec criteria. Sys collate res. Decr probs of interp due to var in samp. Quantif eff sizes and their uncert as pooled estim- pooling decr CI I incr cert of interp eff size.
-qual criteria- should have formal protocol specif- compilation of compl set of studies, I’d comm variab/categ defin, stand data extrac, anal allowing for sources var. qual of studies used.
-sys Rev and meta anal- YS Rev not necess incl meta anal if eg clinic heterogen too great eg too few papers/lot diffs btw studies= can’t comb.
OR= odds surv with tx/odds w/o. CI.
Pooled estim OR=OR for all studies in meta and comb using stats prog. Studies weighted acc to size and uncert of OR (smaller ef=greater weight).
LO4: interp forest plot from a meta anal.
-fixed eff mod= give each study weight. % weight high if large trial and stat sig. Diamond=overall OR.
LO5: list comm diffic said in meta anal and sys Rev.
-heterogen btw studies-
Mod for var- fixed effmod vs rand effs.
Anal var- sub grp anal.
Var qual of studies.
Public bias in study selec.
Heterogen invest in meta- ideally sim design, pp prof, tx/expos, outc meas, stat anal. BUT none identic.
-mod for var- 2 appr calc pooled estim OR+95% CI-
Fixed eff- ass studies estim same size eff. Plot res and rand err bar. Assume all studies trying to meas same true eff- any var from it is rand err.
Rand effs- ass studies estimating sim, not same, eff size. No sing true eff, each study trying find own trial specif eff, all Sep ones distrib on plot with diff AV lines, can comb to calc mean eff of diff trial effs. Is distrib of true effs. Rand effs mod often= more even weighting- each meas own eff, not outweigh. CI comm wider as more uncert as not 1 true eff.
-fixed eff vs rand- pt estim eg OR often simil. 95% CI often wider for rand and more weighting given to smaller studies. Hyp for heterogen- low stat pow to detec heterogen, often use 10% sig. LOT debate which better.
-meta- contin outc eg LT decr salt diet.
Mean change from baseline BP comp to contr. Test for heterogen: null hyp is no heterogen. Diamond= overall test for no diff.
-heterogen- anal variat.
Rand eff mod can acc for var NOT expl it. Sub grp anal expos, may insight into eff of tx/expos, is eff more amplif in cert grps?
2 types sub grp anal:
Stratification by study chars- subsets of whole studies defined by eg study design eg length foll up, or pp prof eg selec criteria.
Stratif by pp prof- where data anal by types pp eg age, sex. Greater stat pow than indiv studies but data often unavailable.
-var qual of study-
Due to poor design, protocol, implementation etc, harder to assess.
Some studies more prone to bias/confound. Least to most suscep- RCT, NRCT, coh, CC.
2 apprs used-
Def basic qual stand for study inclus.
Score each study for qual then: (somet thresh cut off, somet use in mod) incorp qual score into weighting. Use sub grp anal to explore diffs. Meta regress anal Ie weighted Lin regress of eff size Ag qual.
-assess qual- for RCTs many scales-
Main compons- alloc meth eg randomisation, blinding, pt attrition eg und 10% and ITT. Approp stat anal. Who asses qual- more than one assessor. Handling disagreements. Somet diffic to blind assesors. Excess complex stat anal sigs bad study.
Which used for the forest plot?
LO6: desc eff of public bias.
- stat sig or fav res more pub and subm. Partic small studies. = when Rev small studies= fav eff. = bias study selec. Need detec in pooling.
- meths ID- check meta protocol for ID meth-should incl search unpublished studies. Plot id’d studies vs their size= funnel. Stat test for public bias us weak. OR X acis, 1/stand err y axis. Likely overestim true eff.
- funnel interp- plot study size vs eff. Bal=no bias. Smaller studies vary more from cent eff size. Bias if few small studies with small/neg eff.
- critic appraisal of sys Rev and meta- NHS CASP- is valid, what res, will res help loc.
- sources sys revs-cochrane lib, NHS cent for revs and dissem, NIHR Health teach ass prog.
LO1: expl role of ev in clinic prac and lims of using ev from indiv studies.
-services and intervenes based on best AV ev- rigour research.
Prim study eg RCT. Sec lit Rev-
Narrative/expert Rev- implicit assumps, opaque meth, not repro so biased and subj. disc and summ lit on topic w/o gen pooled figs by meta anal. =comp overview, not address Q. Not rep meth/how reflec studies selec. NOT sys Rev.
Sys Rev- explicit assumps, transparent meth, repro so unbiased, obj.
Synth of med research on subject. Thorough meths search for and incl all poss research. Only relev, us min qual.
