評讀系統文獻回顧 Steps in conducting a Systematic …

1 Systematic Reviews ( ) Stepsin conducting a SystematicReview Systematic ReviewandMeta-analysis Critical Appraisalof theSystematic ReviewWhy Systematic reviews andmeta-analysis? Too much information Increased power Less expensive and quicker toconduct then new studies Limits bias Generalise resultsSteps in conducting a SystematicReview Formulationof a question Location andselectionof studies Assessment of qualityof trials Data collection andextraction Data analysis Presenting andinterpretingresultsKey components of a question Clearly defined question that shouldspecify(PICOS): Types ofparticipants Types ofinterventions Comparators Types ofoutcomes Types of studiesTypes of the disease or conditionsthat are of of interest must the SettingTypes of Interventions Clear description control groups,blinding Explanatory trials Determine if an intervention is beneficial in anideal situation Seek to maximiseinternal validity Pragmatic trials Determine if an intervention is beneficial inrealclinical practice Seek to maximiseexternal validityComparators The next step involves defining thecomparison interventions orcomparators a placebo nothing some other treatmentTypes of outcomes Determine the outcomes of interest Howare they measured?

2 Whenare they measured? Whichare important?Types of Studies RCTs Psuedo-RCTs RCTs are unlikely to be done Ethical reasons/adverse effects Financial reasons Lack of trials Important finding gaps! Guides future researchSearching for ExperimentalStudies Can search by study design Use terms related to methodology ofinterest randomi$ed controlled trial clinical trial prospective, follow up, random etcSearching for Trials Effective search strategy Sensitive find as many studies aspossible Minimise bias think aboutfinding/including studies that are notin major databases Efficient look in the place youexpect to have highest yieldSearch Strategy Features of search strategy Sensitivity ability to identify all the relevantstudies Specificity ability to exclude irrelevant studies,also known as precision Inverse relationship between sensitivity andspecificity means that a large number ofarticles retrieved may not be relevant to thereview question High sensitivity will tend to have low specificitySearch Strategy Steps Initial Search initial search of MEDLINE, CINAHL.

3 Followedby analysis of text words in the title andabstract Second Search all identified key words and index terms acrossall databases Third Search references of identified studies andunpublished studiesDocumenting your Search Important!Keep an accurate record of thesearch and how it was performed Same requirement as primary research Duplicate the search strategy Avoid having to repeat searches Where and how to document? Should be detailed in the search strategy section of RevMan Reference management software ReferenceManager, ProCite, or EndnoteQuestion and ProtocolDevelopment Telephone interventions for symptommanagement in adults with cancer(Protocol) 1 Multifactorial assessment and targetedintervention for preventing falls andinjuries among older people incommunity and emergency caresettings: Systematic review and meta-analysis Pulmonary rehabilitation followingexacerbations of chronic obstructivepulmonary disease(Review) Systematic ReviewandMeta-analysis 2 3 Systematic Review Literature Review Systematic ReviewSystematic Review Appraisal SheetUniversity of Oxford, 2005 Are the results of the review valid?

4 What question (PICO) did the systematicreview address? PICO What is best? The main question beingaddressed should be clearlystated. The exposure, such as atherapy or diagnostic test, andthe outcome(s) of interest willoften be expressed in terms ofa simple do I find theinformation? TheTitle, Abstractor finalparagraph of theIntroductionshould clearly state thequestion. If you still cannot ascertainwhat the focused question isafter reading these sections,search for another paper!Yes No Unclear F-Is it unlikely that important, relevantstudies were missed? What is best? The starting point for comprehensivesearch for all relevant studies is themajor bibliographic databases ( ,Medline, Cochrane, EMBASE, etc) butshould also include a search ofreference lists from relevant studies,and contact with experts, particularlyto inquire about unpublished studies. The search should not be limited toEnglish language only.

5 The search strategy should includeboth MESH terms and text do I find theinformation? TheMethodssection should describethe search strategy, including theterms used, in some detail. TheResultssection will outline thenumber of titles and abstractsreviewed, the number of full-textstudies retrieved, and the number ofstudies excluded together with thereasons for exclusion. This information may be presented ina figure or flow No Unclear 1004 references832 referencesScanned Ti/Ab172 duplicates117 studiesretrieved715 do not meetIncl. criteria82 do not meetIncl. criteria35 studies forCritical Appraisal26 studies review6 studies meta analysis20 studies narrative9 excludedstudiesA-Were the criteria used to selectarticles for inclusion appropriate? What is best? The inclusion or exclusion ofstudies in a Systematic reviewshould be clearly defined apriori. The eligibility criteria usedshould specify the patients,interventions or exposures andoutcomes of interest.

6 In many cases the type of studydesign will also be a keycomponent of the do I find theinformation? TheMethodssection shoulddescribe in detail the inclusionand exclusion criteria. Normally, this will include thestudy No Unclear Telephone interventions forsymptom management in adultswith cancerA-Were the included studies sufficientlyvalid for the type of question asked? What is best? The article should describehow the quality of each studywas assessed usingpredetermined qualitycriteria appropriate to thetype of clinical question , randomization, blindingand completeness of follow-upWhere do I find theinformation? TheMethodssectionshould describe theassessment of quality andthe criteria used. TheResultssection shouldprovide information on thequality of the No Unclear T-Were the results similar fromstudy to study? What is best? Ideally, the results of thedifferent studies should besimilar or homogeneous.

7 Ifheterogeneityexists theauthors may estimatewhether the differences aresignificant(chi-square test).Possible reasons for theheterogeneity should do I find theinformation? TheResultssection shouldstate whether the results areheterogeneous and discusspossible reasons. Theforest plotshould showthe results of the chi-squaretest for heterogeneity and ifdiscuss reasons forheterogeneity, if No Unclear What were the results? A Systematic review provides a summary of the data fromthe results of a number of individual studies. If theresults of the individual studies are similar, a statisticalmethod (calledmeta-analysis) is used to combine theresults from the individual studies and an overallsummary estimate is calculated. The meta-analysis gives weighted values to each of theindividual studies according to their size. The individualresults of the studies need to be expressed in a standardway, such asrelative risk, odds ratio or mean differencebetween the groups.

8 Results are traditionally displayed in a figure, like the onebelow, called aforest Data: Effect of Treatment orExposure Dichotomous Effect/no effect Present/absent Continuous Interval or ratio leveldata BP, HR, weight, etcForest plotFavours treatmentFavours controlNo effectResults of different studies combined The Effect of Shift Rotation onEmployee Cortisol Profile, SleepQuality, Fatigue, and Attention Level:ASystematic ReviewSystematic Reviewbut NoMeta-analysis 4 --SR Comparing groups-2 2 tableLungcancerNo LungcancerTotal(by group)Smoking175168No smoking95564 Total(by event)26106132 Meta-analysis of Dichotomous Data Risk ratio (RR)=risk of event in intervention grouprisk of event in control group Risk ratio=17/68= the risk of developing lung cancer because of smoking was179% of the risk without smoking the risk of lung cancer with smoking is 79% highercompared to no smoking smoking increased the risk by 79%Meta-analysis of Dichotomous Data Odds ratio (OR)

9 =odds of an event in intervention groupodds of an event in control group Odds ratio=17/51= smoking doubled the odds of developing lung cancer the odds of developing lung cancer with smoking is 106% highercompared to no smoking smoking increased the odds by 106%Confidence Intervals Gives an estimated range of values which islikely to include an unknown populationparameter Usually calculated so that this percentage is95%, but we can produce 90%, 99%, ,confidence intervals for the unknownparameter Can tell us about statistical significance butalso tell us about how large or small the realeffect might be precision!Heterogeneity Variation between the results of a set of studies To determine whether significant heterogeneityexists look for theP value for the test of heterogeneity Calculated withForest Plot Quantifying heterogeneity I2describes the proportion of total variation acrossstudies due to heterogeneity rather than chance low (and negative) values of I2indicate no, or littleheterogeneity larger values of I2show increasing heterogeneityCheck for Heterogeneity How to look for heterogeneity?

10 Visual Forest plot: do confidence intervals of studiesoverlap with each other and the summary effect? Statistical tests: Chi-square testfor heterogeneity (Cochran Q test) Tests whether the individual effects are farther awayfrom the common effect, beyond what is expected bychance Has poor power P-value < significant heterogeneityQuestion & Discussio