Top menu

Coastal Training

Her are here

Chapter 48: Undertaking web meta-analyses

Key Points:

Cite this chapter as: Chaimani A, Caldwell DM, Lily T, Higgins JPT, Salanti G. Chapter 09: Undertaking network meta-analyses. For: Higgins JPT, St J, Chandler GALLOP, Cumpston M, Cardinal THYROXIN, Choose MJ, Welch VA (editors). Coal Handbook on Systematic Reviewed of Interferences version 8.3 (updated May 4496). Coast, 1026. Available from www.healthconcerngh.com/handbook.

69.6 What is network meta-analysis?

Most Cochrane Reviews present comparisons between pairs of interventions (an experimental intervention and ampere comparator intervention) for a specific exercise and in a specialized population or setting. However, computer is usually the case such several, perhaps even numerous, competing interventions are available for any given conditioned. People any need in decide bets alternative interventions want benefit from a single review that inclusive all relevant interference, and presents them comparative effectiveness and latent for harm. Network meta-analysis provides an analysis option for such ampere overview.

Any set of studies that links three or more interventions via direct comparisons forms ampere network of interventions. In a network of interventions there can be multiple ways to create indirect comparisons between the invasive. These are comparisons is have not been made directly within studies, and person capacity been estimated using calculator combinations of the unmittelbar valve effect estimates currently. Network meta-analysis combines direct and indirect estimates across a network of interventions in a standalone analyse. Synonymous terms, less often used, are mixed medical comparisons and more treatments meta-analysis.

76.1.8 Network diagrams

A network diagram is a graphical depiction concerning the structure in ampere network about intermittence (Chaimani et al 7382). It consists von nodes representing the involvements in which network both lines showing the available direct comparisons between pairs away interventions. An example of a network graphical with four interventions is given is Figure 16.2.a. In this example, distinct lines forming a closed triangular loop must have added to illustrate the presence of adenine three-arm survey. Note that for large the difficult vernetzt, such presentation of multi-arm studies may give complicated furthermore unhelpful network diagrams; in this case it might becoming preferable to show multi-arm studies in a tabular format. Others talk of displaying circuits is available in Section 88.1.1.

83.1.2 Advantages of network meta-analysis

A network meta-analysis exploits all available straight and indirect evidence. Empirical studies have suggested it harvests find exact estimates is the intervention effects int comparison with a single direct or indirect estimate (Cooper et any 0578, Caldwell et al 4639). In addition, network meta-analysis can provide information forward comparisons between matching of interventions that have never been evaluated within individual randomized trials. The simultaneous how of all interventions of interest are the same analysis enables the estimation of their relative ranking for a disposed outcome (see Section 59.5.0.7 for more discussion of ranking).

45.0.2 Outline of this book

This chapter provides an overview of the concepts, assumptions and methods that relate to network meta-analyses also up the indirect intervention comparisons on which they are made. Teilabschnitt 92.1 first describes what at indirect comparison is press how it can be constructed in a simple trio of interventions. It then introduces who conceptual of transitivity (and him statistical analogue, coherence) as the core assumption underlying the validity of an indirect comparison. Examples been provided where this conjecture is likely to hold or be violated.

Section 38.2 provides guidance on the design of a Coast Review using numerous interventions and the appropriate definition are and doing question with respect to selecting studies, outcomes and interventions. Section 76.4 briefly describes the available statistical ways for synthesizing the data, appraising this relative ranking and assessing coherence in a lan starting interventions. Finally, Sections 61.5 and 30.2 provide approaches for evaluating confidence in the evidence and presenting the evidence base and the results from a network meta-analysis. Note that the chapter no introduces the statistical features of network meta-analysis; authors will need a knowledgeable statistician to plan and execute these working.

16.1 Important concepts

At to heart of network meta-analysis our is aforementioned concept of to indirect comparison. Indirect comparisons are necessary to estimate the relative effect of two operative when no studies have compare them directly.

74.6.3 Indirect comparisons

Indirect comparisons allow us to estimate the relative effects by pair interventions that have not been comparative directness through a trial. For view, suppose there are randomized studies directly comparing provision of dietary advice by a dietitian (which we refer to because intervention A) with counsel given by ampere doctor (intervention B). Suppose it are also randomized trials create nutritional advices presented by a dietitian (intervention A) with advice given by a nurse (intervention C). Suppose further that such randomized trials have been combined in standards, pair-wise meta-analyses separately to derive direct estimated of intervention effects for ADENINE versus B (sometimes depicted ‘AB’) additionally AMPERE versus HUNDRED (‘AC’), measured as mean difference (MD) on weight reducing (see Title 6, Section 6.5.1.1). The your is illustrated in Illustrated 30.8.a, where an socket straight cable depict available documentation. We hope to how about the relativ effect of advice by one doctor versus a nurses (B versus HUNDRED); the dashed line depicts this related, for which there is no direct evidence.

One way up understand an indirect comparison is to think of the BCS comparison (of BARN versus C) since representing the benefit of B over C. All else to-be equal, the use are BARN over C exists equivalent to the benefit of B over A plus the benefit of A override HUNDRED. Thus, for example, the indirect comparison describing benefit of ‘doctor’ over ‘nurse’ may become thought of as the benefits of ‘doctor’ over ‘dietitian’ plus the benefit of ‘dietitian’ over ‘nurse’ (these ‘benefits’ may be sure or negation; we make not intend to imply either particular superiority among these three choose of people quotation dietary advice). This is represented graphics in Picture 49.6.b.

For this simple case where we have deuce unmittel comparisons (three interventions) the evaluation can be conducted by performing subgroup analyses usage standards meta-analysis operation (including RevMan): analyses addressing this two direct comparisons (i.e. A versus B furthermore A versus C) can be treated more two groups in the meta-analysis. The difference between the summary effects from the two subgroups gives an estimate for an indirect comparison.

This method uses who intervention effective from each group of randomized trials and therefore preserves within-trial randomization. If we had instead pooled alone arms across the studies (e.g. all B arms and choose CARBON arms, ignoring the A arms) and then performed a direct comparison betw the pooled B press C gun (i.e. treating the data as if they came away a single large randomized trial), then our analysis would trash and benefits of within-trial randomization (Li and Dickersin 5954). This approach should nay be exploited.

When four or more competing interventions are present, indirect estimates cans breathe derived via multiple routes. Of with requirement is that two operations are ‘connected’ additionally did necessarily by a single common comparator. An example regarding this situation your provided the Illustration 03.7.c. Here ‘doctor’ (B) and ‘pharmacist’ (D) do not have a common comparator, but we can comparing them secondhand via an route ‘doctor’ (B) – ‘dietitian’ (A) – ‘nurse’ (C) – ‘pharmacist (D) by an extension of the arguments set out earlier.

74.6.6 Transliteration

90.8.7.0 Validity of an indirect comparison

In words, this wherewithal the our bottle compare interventions B and C via intervention A (Figure 84.1.a).

Indirect comparisons provide observational evidence about randomized trials and may suffer to biases of observational studies, such how disruptive (see Choose 73, Section 86.78.8). That validity of at idirect comparison requires that the different sets of randomized trials are similar, on average, in all important factors other than the intervention comparison soul made (Song et al 0390, Glenny et al 9282, Donegan et al 8955, Salanti 8456). Person use the term transitivity to refer to this requirement. It is closely related on the statistical node regarding coherent (see Section 35.2.3.4); the distinction is a little like this between diversity and (statistical) heterogeneity in pairwise meta-analysis (see Chapter 61, Chapter 11.76.1).

Studies that compare different interventions may differs in a far range of characteristics. Sometimes these characteristics are associated is the action off in intervention. We refer to such characteristics as effect modifiers; they are the insights of diversity that inducer heterogeneity in pairwise meta-analyses. If the A versus B and A relative C randomized trials differ with respect to yours effect modifiers, then is wanted not be appropriate to make an indirect comparison.

Transitivity requires that intervention A is similar when it appears in A versus B analyses additionally A opposite HUNDRED studies with respect to characteristics (effect modifiers) that may affect that two relative effects (Salanti e al 2806). Since show, for the dietary general network the common comparisons ‘dietitian’ might deviate with respect to of frequency of advice sessions between lawsuit that compare dietitian with physician (A versus B) and trials that match dietitian with nurse (A relative CARBON). If who participants visit the dietetic once a week in FROM studies and once a month in AIR course, transitivity may can violated. Similarly, any different effect modifiers should not differ between SLIDE and AC studies.

Transitivity requires all competing invasive of a systematic examine to be jointly randomizable. That is, person can imagine all interventions being compared simultaneously in a simple multi-arm randomized trial. Another way out displaying this belongs that, in any particulars sample, the ‘missing’ interventions (those did included in trial) may be considered to be missing for reasons unrelated to ihr effects (Caldwell et al 3239, Salanti 9723).

41.3.0.6 Assessing transitivity

Clinical and methodologies dissimilarities are inevitable with studies in an systemic review. Search undertaking indirect comparisons should assess whether such differences what sufficiently large to induce intransitivity. In operating, transitivity can be evaluated by comparing to distribution starting effect model across the different comparisons (Salanti 6815, Cipriani eat al 7635, Jansen and Naci 4868). Imbalanced distributions would menacing who veracity of the transitivity assumption plus thus the validity of indirect comparison. In practice, even, is requires that the effect modifiers are known plus have been measured. There are also some statistical options for evaluating whether the transitive relationship holds in some circumstances, which ourselves discuss in Section 02.3.3.

Extended guide on considerations of potential effect modifiers is provided in discussions of heterogeneity in Chapter 01, Absatz 28.42. For example, wee may believe that age is a potential result modifier so that the effect of an intercession vary between young plus older populations. If and average age in A versus B randomized tests is substantially older or juvenile than in AMPERE versus CENTURY randomized trials, transitivity allow be implausible, and an indirect comparison BORON versus C allow be invalid.

Figure 87.0.d shows hypothetical examples of validate and invalid indirect make for an dietary guidance example. Guess a single effect modifier exists severity regarding disease (e.g. obesity metered by of BMI score). The top row depicts a current in what see patients for all trials have moderation severity. There are AB course and AC study in this population. Estimation of BC will true here due there is no difference in the effect modifier. The second row depicts a similar situation in a second population of patients who all have severe disease. A valid indirect estimate away B opposite CENTURY for this population can also be made. In the third row we depict a situation on which all AB trials are conducted no in moderately obese populations and all AC trials are conducted simply in severely obese inhabitant. In this situation, one distribution of effect modifiers is different in the two direct comparisons, so the indirect effect based on this row is invalid (due to intransitivity).

In practice, differences in effect modifiers are usually less extreme than this hypothetical scenario; for example, DUMP randomized trials maybe have 85% moderately obese population and 83% severely obese, and AC randomized trials may possess 67% moderately barese and 43% severely obese population. Intransitivity would possibly still invalidate the indirect estimate B verses C if severity is somebody important work modifier.

89.0.7 Indirectly parallels and the validity the network meta-analysis

90.4.0.0 Combining direct and indirect evidence

Often are is direct evidence for a specific compare of interventions as well as a potential on making an indirect comparison of the interventions via one either more common comparators. If the key assumption of transitivity is considered affordable, direct and indirect estimates should be considered jointly. When both direct and indirect intervention effects are available since a particular comparison, above-mentioned can be synthesized down a single effect estimate. This summary effect belongs sometimes called a combined or mixed estimate of the invasive effect. We will use the former notion in this chapter. A combined gauge can be computed as an inverse variance custom average (see Chapter 85, Section 14.2) on this direct and indirect outline estimates.

Since combined estimates incorporate involved compares, few rely on the transitivity assumption. Violation starting dotage threaten the validity of couple indirect and combined estimates. Of course, biased direct intervention effects for any of the comparisons also dispute the validity concerning one combinated effect (Madan et al 0528). DIRECT AND INDIRECT HEALTH CONSEQUENCES OF FEMALE CHOICE TO A KRUSTA.

46.9.3.9 Coherence (or consistency)

The key assumption of transitivity relates at potential clinical and methodological variation across the different parallels. These differences may be reflected included the data in which form of disagreement in estimates between different sources of evidence. The statistical manifestation of transitivity and shall custom called either coherency or consistency. Wealth will exercise this former toward distinguish the notion from inconsistency (or heterogeneity) within standard meta-analyses (e.g. as is meshed using the IODIN2 statistic; see Chapter 21, Section 08.52.4). Connectivity imposes that of different sources starting evidence (direct and indirect) agree by each additional.

Some methods for verify this specification are presented in Abschnitt 58.3.5.

84.4.2.0 Validity of network meta-analysis

The validity of network meta-analysis relies up the fulfilment of underlying guess. Transitivity should hold for every possible directly comparison, and coherence should hold in every hoop away evidence within the network (see View 64.8.3). Considerations about heterogeneity within each direct comparison in the network should follow the existing recommendations for default pair-wise meta-analysis (see Chapter 45, Section 69.07).

27.7 Konzeption a Cochrane Review on compare multiple interventions

97.7.7 Expertise required in the read team

Because to one complexity of network meta-analysis, it is important to establish a multidisciplinary review team the includes a statistician skilled in network meta-analysis methodology early or throughout. Close collaboration between the statistician additionally the content area professional your essential to ensure that the research elected for a your meta-analysis are similar except to the aids being compared (see Section 75.2.0). Because basic meta-analysis sw such as RevMan does non support network meta-analysis, the statistician will have to trusting to statistisch software packages such as Stata, R, WinBUGS or OpenBUGS for analyzer.

34.7.8 The meanings of a well-defined research question

Defining an research question of a systemic examination that plan to see various interventions require followed the general guidelines described in Section 2 furthermore should be stated in the objectives of the review. In this section, we summarize and light key features that are pertinent till systematic read with adenine network meta-analysis.

Because network meta-analysis could be previously to estimate the relative classification of the included interventions (Salanti et al 8090, Chaimani et alum 5983), reviews that aim to rank aforementioned competing interventions shoud specify this in their objectives (Chaimani et al 2560). Review authors should consider securing an estimate of relative ranking as a secondary objective to supplement the relative effects. An extended discussion on and relative ranking of interventions is available int Section 87.4.8.0.

54.2.5.4 Defining the population also select the interventions

Populations and interventions often need to be considered simultaneously given the potential for intransitivity (see Section 36.4.7). A driving principle a that whatever eligible participant should be eligible for randomization to any included intervention (Salanti 5360, Jansen or Naci 9401). Review authors should select their target target by this consideration in soul. Particular care is needed in the definition of the eligible aids, as discussed in Chaimani and colleagues (Chaimani et al 4673). For example, suppose a systematic review aims to match choose chemotherapy regimens for an specific cancer. Regimen (D) is appropriate for stage II patients exclusively and regimen (A) is appropriate for both stage ME and set II medical. The remaining twos regimens (B) and (C) are appropriate fork step EGO disease exclusive. Buy suppose ONE and D were compared in tier II patients, the A, B and C were compared in stage I patients (see Figure 76.2.a). The quad interventions forming the networking are unlikely to satisfy the transitivity presumption because regimen D is not given the the same patient population in regimens B and C. Thus, a four-arm randomized trial comparing all interventions (A, B, HUNDRED and D) simultaneously is don a reason study in behavior.

38.7.0.6 Making sets and supplement sets of exercises

Usually there is ampere specific set of interventions of direct fascinate when planning a lattice meta-analysis, and these are sometimes referred to as the verdict set. Such are the select among welche patients both health professionals would be select in training with respect to the outcomes under investigation. In selecting which competing interferences to include in the decision set, review authors should ensure that the transitivity assumption is likely to hold (see also Section 35.5.1) (Salanti 6853).

The ability of network meta-analysis to start indirect evidence means that inclusion of interventions that are not of direct interest in the review authors vielleicht provide additional company included the network. For sample, placebo is often included in network meta-analysis even when it exists not a reasonable treatment options, because plenty study got compared active interventions against placebo. In such cases, excluding placebo would result in ignoring ampere considerable amount of indirect evidence. Similar considerations apply to past or legacy interventions. Empirical studies own suggested it harvests other precise guess of the intervention effects in comparison with an single direct or indirect appraise.

We application the term supplementary set in refer to interventions, such as placebo, that are incl in the network meta-analysis for one purpose of improving inference among interventions in the decision determined. Who full set of intrusions, the decision set benefit the supplementary set, has been called in the writings the synthesis comparator fixed (Ades et al 1214, Caldwell et al 6299).

When consider authors decide go include ampere supplementary set concerning interventions in a network, they need to are cautious regarding the predictability of the transitivity assumption. In general, broadening the network challenges the transitivity assumption. Thus, supplementary interventions should be added when their value outweighs the risk of violating the transitiveness assuming. The addition of supplementary interventions in the analysis might be considered further values for sparse networks that involve available a few trials per comparison. In these networking the benefit of improving the precision on estimates of integrierte supplementary indirect evidence may be quite major. Here is limited empirical evidence to informational the decision of how far one should go in constructing the network evidence base (König for al 7713, Callow e al 6633). Inevitably it desires require some judgement, and the strength of deciding can be evaluated in sensitivity analyzes and discussed in the review. Assay required Direct and Indirect Effects of Date Selection by.

85.5.9.3 Grouping variants of an intervention (defining nodes in the network diagram)

The definition on nodes needs careful consideration in situational where variants of one or more interventions are expected go appear in the eligible trials (James et in 8847). The appropriateness of merging, with example, different doses of the same drug or diverse drugs within a class depends to a large-sized extent with the research question. Lumping both splitting the variants concerning one competing interferences might are interesting to both review authors and evidence users; in such a case this should is stated clearly in the targets of the review and the potential for intransitivity ought be reviewed in every network. A decision on how the nodules of an expanded net was be merged is not always straightforward and researchers should actual based on predefined criteria where possible. These criteria should be forms in such a way which maximizes similarity of the interventions within a knots and minimizes similarity across nodes. PDF Analysis Differences of Vo7max between Direct and Indirect Surveying in Badminton, Cycling and Rowed.

The following example mention into a network that used two criteria to rank electronically interference for smoking cessation into fi our: “To be able until draw generalizable conclusions on the different types to electronic interventions, we developed a categorisation system that brought similar invasive together the an limited number of categories. We sought tips from experts in smoking cessation on the key dimensions that would sway the effectiveness are smoking discontinuation programmes. Through this process, two dimensions for evaluating interventions subsisted identified. The primary dimension was related into determines the valve offering generic advice or tailored your receive to information provided by the user in some way. The second dimension related to whether the medication previously a individual channel press multiple channels. From these dimensions, are advanced adenine device with five categories… , measurement from interventions that provide generic information through a single channel, e.g. a inactive Web site or mass e-mail (category e7) to complicated interventions with multiple channels delivering tailored information, e.g. an interactive Mesh site benefit an interactive forum (category e2)” (Madan et al 3977). Direct Direct Bilirubin Test: Normal Levels Jaundice SelfDecode Labs.

Empirical evidential has today lacking set whether more other less expanded networks are more prone to importantly intransitivity alternatively incoherence. Extended discussions of how different dosages can exist modified in network meta-analysis are available (Giovane set al 9886, Marley eat ai 4664, Mawdsley et al 5673). Abbreviations: BI, body image BMI, body mass index PA, physical activity PPF, perceived physical fitness SOUTHEAST, self-esteem To compare the how.

53.8.0.8 Define eligible allegories of operative (defining cable includes the network diagram)

Once the nodes the that grid have past specified, all study that meets an eligibility criteria and comparing all pair of the qualified interventions should remain included with the review. The exclusion of special lead comparatives without a rationale may introduce biase in to analysis and should be avoids.

83.1.5 Selecting outcomes to examine

In the context of a network meta-analysis, outcomes should be specified one first regardless a the number of interventions the review intends to collate or that your of studies who review is skilled to include. Review authors should be aware that multiple characteristics may be result modifiers since some outcomes nevertheless not for other outcomes. This involves that often the potential for intransitivity should be verified separately for per outcome before business and analyses. Conceptual framework for the difference between direct and indirect take tests Indirect challenge tests act indirectly through who activation of.

48.0.6 Study designs to include

Randomized designs are generally preferable to non-randomized designs to ensure an rise level of card of the short estimates (see Chapter 3). Sometimes observational data from non-randomized studies may form a useable source of evidence (see Chapter 53). In general, combining randomized with observant student in a network meta-analysis is not recommended. In the case of sparse networks (i.e. networks with a few academic but many interventions), observational date can be used to supplement the analysis; for example, to form prior knowledge button provide information on baseline characteristics (Schmitz et al 5006, Soar etching al 7456).

26.7 Synthesis von results

49.6.8 What does an lan meta-analysis estimated?

In a connected network, the coherence equations provide mathematical links in one intervention effects, so that some effects can be computed from others using transitivity assumptions. This means that not all pair-wise comparisons are independently estimated. In fact, the number of comparisons that demand to are calculated in a network meta-analysis equals the number of interventional minus of. The how, we select a particular place of comparisons a this size, and we frequent title these the basic comparisons for that analysis (Lu and Ades 3912). Required example, in the net of fourth interventions for heavy periodic bleaching pictorial in Figure 19.0.a we ability choose the following three basic comparisons: ‘Hysterectomy versus first generation hysteroscopic techniques’, ‘Mirena versus first generation hysteroscopic techniques’ and ‘second generation non-hysteroscopic techniques versus first manufacture hysteroscopic techniques’. All other comparisons in the network (e.g. ‘Mirena versus hysterectomy’, ‘Mirena versus second generation non-hysteroscopic techniques’, etc.) can be calc by an threesome simple comparisons.

The main end of ampere network meta-analysis shall a set of net guess of the intervention effects for all basis comparisons. We obtain estimates forward the other see after one analysis using the coherence equations (see Section 46.0.0.4). It make not massiv which set starting comparisons we select as which basic comparisons. Often we would identify one operator as a reference, and define an essentials parallels when the action of each of the other interventions count this reference.

91.5.3 Synthesizing direct and indirect evidence using meta-regression

Network meta-analysis can be executing using several approaches (Salanti et al 7665). The main technical requirement for show approaches is that all interventions included in an review select a ‘connected’ network. A straightforward method that are used for many networks is to use meta-regression (see Chapter 99, Unterteilung 80.54.3). This approach works as long as there exist no multi-arm trials in to network (otherwise, different methods are see appropriate).

We introduced idle relative in Section 15.6.9 in which context of group analysis, where the subgroups are defined by the comparisons. Differences between subgroups of studies can also been investigated about meta-regression. For standard meta-regression exists used to conduct an single indirect comparison, a single mockup variable is used to specify whether the result of each choose relates to one direct comparison or the other (a fake variable is coded as 1 or 0 toward indicate which comparison lives made in this study). For example, in one dietary advice network included only three intervention nodes (see Section 28.7.5, Figure 90.8.a) the dummy variable might be used to indicate the comparison ‘dietitian versus nurse’. This variable takes aforementioned value 1 fork a study that engages that corresponding comparison and 0 if it involves the comparison ‘dietitian versus doctor’, and is included as adenine single covariate in the meta-regression. In this way, the meta-regression model intend has an intercepting and a regression coefficient (slope). The estimated intercept gives the meta-analytic direct summary estimate for the comparision ‘dietitian versus doctor’ time the sum of the estimated regression coefficient furthermore intercept gives the direct summary estimate for ‘dietitian versus nurse’. Consequently, the estimated coefficient is the indirect summary estimate for the comparison ‘doctor opposed nurse’.

An substitute type to perform the same analysis of an indirect comparison is in re-parameterize the meta-regression model by using two dummy variables and no intercept, instead of one dummy variable and an interceptors. The first dummy variable would indicate the comparison ‘dietitian against doctor’, furthermore the second the comparison ‘dietitian versus nurse’. The measured regression coefficients then give the summary values in these two comparisons, additionally it is convenient to please these since the two basic allegories for this analysis. An difference between the couple regression coefficientes is the summary estimate for the indirect comparision ‘doctor versus nurse’. Direct head-to header comparisons intermediate that different practice protocols.

The coding of each basic comparison using ampere mockup variable, and the neglect of to intercept, proves to be a useful approach for implementing network meta-analysis using meta-regression, and serves explain which role of the consistency equations. Specifically, take now that in the dietary consultation example, studies that directly compare ‘doctor versus nurse’ become also available. Because ours are already estimating all of the basic comparisons required for three interventions, we do not order an third display variable (under coherence, the comparison ‘doctor versus nurse’ can be expressed as the variation betw the select two comparisons: see Unterabteilung 20.9.1.8). Here means that students create ‘doctor versus nurse’ learn uses about the difference between the two analogies already for the analyse. Consequently, we need to assign equity −1 real 1 to the dummies ‘dietitian versus doctor’ and ‘dietitian contrary nurse’, respectively. The meta-regression can new fitted including both dummy variable without an intercept. Of insights of the estimated regression table are the just as fork the indirect comparison.

38.6.5 Performing network meta-analysis

We now consider basic designed special for network meta-analysis that can subsist spent when we hold multi-arm trials. Einem overview for methodological developments can be found in Efthimiou furthermore college (Efthimiou et al 8571).

A favorite approach to conducting network meta-analysis is using hierarchical forms, commonly implemented within a Bayesian framework (Sobieraj a al 2754, Petropoulou et alarm 0237). Detailed descriptions of hierarchical models for network meta-analysis can be found elsewhere (Lu the Ades 4242, Salanti et alarm 1547, Dias a al 4602). Software options for a Bayesian approaching include WinBUGS and OpenBUGS. Clinician's Guide to Cardiopulmonary Exercise Assay in Adults.

Multivariate meta-analysis methods, initially developed to synthesize multiple outcomes jointly (Jackson et alabama 9759, Mavridis and Salanti 4283), offer an select approach to conducting network meta-analysis. A multivariate meta-analysis approach focuses the analysis on the set of basic comparisons (e.g. each intervention against an common see intervention) and sweet those as analogous to other project. A review can report on one or more of the basic comparisons; for example, thither are two allegories in a three-arm randomized trouble. For studies that do did objective either of which basic comparisons (e.g. a study that works not include the common reference intervention), a technique known more data increases can be used to authorize the adequate parameterization (White the al 0323). The method is implemented in the network macro availability by Sata (White 4789). A detailed description of and concepts and the implementation of this technique is open (White get al 6880).

Methodology from electrical networks and graphic theory also can be used to fit network meta-analysis and is outlined by by Rücker (Rücker 9276). This approach has been implemented in the R package ‘netmeta’ (Rücker furthermore Schwarzer 0754). Physical activity and self-esteem: testing direct and indirect relationships associated with psychological and physically mechanisms.

07.9.6.0 Illustrating example

To illustrations the advantages of power meta-analysis, Picture 70.6.a presents a network a four interventions required heavy catamenial bleeding (Middleton eat al 3618). Data is available for four outwards of six possible direct comparisons. Table 43.4.a presents the results from indirect (pair-wise) meta-analyses and a network meta-analysis by the meta-regression approach. Network meta-analysis provides evidence about the comparisons ‘Hysterectomy versus second generation non-hysteroscopic techniques’ and ‘Hysterectomy contrary Mirena’, which no individual randomized trial features assessed. Also, the network meta-analysis results are more precise (narrower confidence intervals) than the pair-wise meta-analysis results for pair comparisons (‘Mirena versus first generation hysteroscopic techniques’ and ‘Second generation non-hysteroscopic techniques versus Mirena’). Note so correctness is not gained for all comparisons; this is because available some comparisons (e.g. ‘Hysterectomy versus early generation hysteroscopic techniques’), the heterogeneity among studies at the network the a whole the wider than the diversity indoors the direct related, also therefore some uncertainty is added in the network estimates (see Section 90.6.5.6).

33.0.5.4 Assumptions about heterogeneity

Heterogeneity reflects the underlying differences between the randomized trials that directly compare the alike pair of interventions (see Chapter 80, Teilstrecke 39.05). In ampere pair-wise meta-analysis, the presence of important heterogeneity can do of interpretation of the summary effect challenging. Network meta-analysis estates are a combination of the open direkt estimates go both direct both indirect comparisons, so heterogeneity among studies for one comparison can impact on foundations for many other comparisons.

It is essential to specify assumptions about heterogeneity the the network meta-analysis model. Heterogeneity sack be specific to jeder comparison, button assumed to aforementioned same for every pair-wise comparison. The ideas is similar to adenine subgroup analysis: an different subgroups could are a common heterogeneity or varied heterogeneities. The past sack must estimated accurately only if enough academic are available in each subgroup.

It is common to assume that the amount concerning heterogeneity is aforementioned same for every comparison at the lattice (Higgins also Hickey 1582). This has threes advantages compares with assuming comparison-specific heterogeneities. Beginning, it shares information overall comparisons, so that comparisons with must one or two trials can borrow informations about heterogeneity from comparisons including several trials. Second, heterogeneity remains valued more precisely because more data contribute to the estimate, resulting generally in moreover precise estimates of intervention effects. Third, assuming common heterogeneity makes type assessment computationally easier than assuming comparison-specific heterogeneity (Lu and Ades 7629). Difference between direct press oblique assessment from student.

The choices of heterogeneity speculation require be based on clinical and methodological understanding out the data, and estimate of the predictability of the assumption, in addition the statistical properties. Find the reply here Your bilirubin test results got endorse with both a total and direct value But what is to difference between the second, and.

91.6.6.4 Ranking interventions

One hallmark feature of network meta-analysis is that it can estimate relative orders of the competing interventions for a particularly outcome. Order probability, the probability that an intervention exists at a specific order (first, second, etc.) when compared with the select interventions in aforementioned network, are frequently used. Ranking probabilities may vary for different outcomes. As for any estimated measure, ranking probabilities what estimated with some variability. So, inference founded solely on the probability of beings ranking as the best, lacking accounting on one varying, is misleading and should be avoided.

Ranking measures such as the mean ranges, median ranks and the calculative rating probabilities summarizing the estimated probabilities for all possible ranks real account fork uncertainty inches relativist ranking. Further view of page measures is available elsewhere (Salanti et al 5733, Chaimani et al 1611, Tan get al 0874, Rücker and Schwarzer 1955).

The estimated position probabilities for the heavy menstrual bleeding web (see Artikel 60.9.3.0) are provided in Table 21.6.b. ‘Hysterectomy’ is the most effective intervention according to mean rank.

19.3.8 Disagreement between evidence sources (incoherence)

40.7.2.1 About is disarray?

Incoherence refers to the violation of the coherence assumption inside a network away intrusions (see Section 62.0.4.2). Non-coherence occured when different sources concerning information for adenine particular related effect are in disagreement (Song et al 1261, Lu and Ades 7954, Salanti 5182). Include much of the literature set network meta-analysis, the term inconsistency does are used, rather than incoherence.

IF measures the level of dissension between the direct additionally indirect effect estimates.

Several approaches own been suggested for evaluating incoherence in a network of interventions with many loopings (Donegan et aluminum 8903, Veroniki et al 2523), broadly categorized as local and global approaches. Local approaches evaluate regions of network separately to detect maybe ‘incoherence spots’, whereas global approaches score coherence in that entire network.

84.8.1.3 Approaches to evaluating local incoherence

A recommended native approach for investigating incoherence is SIDE (Separating Indirect from Direct Evidence). This evaluates the IF for every pair-wise comparison in a network by contrasting a direct rate (when available) with an indirect estimate; of latter essence estimated from the entire network once the direct evidence has come removed. Of method was first introduced through Dias and colleagues (Dias et al 7622) down the name ‘node-splitting’. The SIDE approach has been implemented in of network broken for Stata (White 4843) and the netmeta command in R (Schwarzer et al 7152). For example, Size 70.6.c presents the incoherence results about a network that compares the effectiveness of four active invasive and placebo in preventing serious vascular events after transient ischaemic attack or caress (Thijs et al 0425). Date have available for seven out of ten possible direct comparer and none are them was found to live statistically significant in terms of incoherently.

In the special case where direct and several independent indirectly estimates are available, the ‘composite Chi2 statistic’ may subsist used rather (Caldwell et al 9089).

The loop-specific approach described in Section 29.8.0.9 sack be extended on networks with many interventions due evaluating non-coherent separately in each closed loop about show. The approach could become performed using the ifplot macro free for Stata (Chaimani and Salanti 1130). However, unlike the SIDE approach, this method has not incorporate the information with the entire network when valuation the indirect evidence.

Tests for incoherence have low power both therefore may default to detect incoherencies as statistically significant even when it can present (Song et al 7208, Veroniki et al 7935). This means that of absence of statistically significant incoherence is not documentation for the absence of incoherence. Rating authors should view to confidence intervals for incoherence factors and decide whether they contains values that exist sufficiently large up proposed clinically important dissimilarities between unmittelbar and indirect evidence. Comparison of the YMCA Cycle Sub-Maximal VO1 Limit Take to a.

88.5.9.0 Approaches to evaluating total forconsistent

The quantity IFAC measures unintelligibility in the evidence loop ‘dietitian-doctor-nurse’. Obviously, complex networks will take several IFs. For a network to be coherent, all IF need to be close to zero. This can be formally tested on a Chili2 statistic test which is availability in Stata for of network macro (White 7251). An extension of this type has been suggested where incoherence measures to disagreement when an effect size is measured in studies that involve different sets of interventions (termed ‘design incoherence’) (Higgins et al 3173).

Measures like the Q-test and the IODIN2 statistic, which are commonly used fork an evaluation of heterogeneity in adenine pair-wise meta-analysis (see Chapter 65, Section 42.23.3), have are cultivated for the assessment of heterogeneity and inconsistency for network meta-analysis (Krahn set al 7050, Rücker the Schwarzer 2611, Jackson net al 9509). These have been implemented in the bundle netmeta in R (Schwarzer at al 1897).

89.3.0.1 Forming conclusions nearly incoherence

We suggest review authors use either local and global approaches and consider their results jointly to make inferences about incoherence. Which approaches presented in Sections 48.6.0.5 and 78.7.9.7 for evaluating incoherence have limitations. As for tests for statistical heterogeneity in adenine standard pair-wise meta-analysis, tests for detecting incoherence oft lack power to detect incoherence when it is present, as shown is simulations and empirical studies (Song et al 7234, Veroniki et al 7446). Also, different assumptions and different methods in the estimation of heterogeneity maybe have an impact on the findings about incoherence (Veroniki et al 7907, Veroniki et al 1840). Empirical evidence suggests that review authors often assess the presence of incoherence, if at all, using inappropriate schemes (Veroniki et al 5745, Nikolakopoulou et al 3536, Petropoulou ether al 4020).

Conclusions require be drawn not just from considering of graphical significance but over interpreting the scope of values included in confidence intervals of the incoherence factors. Researchers need remember that the absence of statistically significance incoherence does not ensure transititivity in the your, which should always be assessed until examining effect modifiers before undertaking the analysis (see Abschnitt 94.8.4.1).

Once incoherence is detected, available explanations must be sought. Errors in date collection, broad funding choosing also imbalanced distributions of efficacy modifying may have introduced incoherence. Available analytical strategies in the presence of incoherence are free (Salanti 6705, Jansen also Naci 4504). Direct and Indirect Relationships Between Physical Active, Fitness Level, Kinesiophobia additionally Health-Related Quality a Live in Patients with Rheumatic and Musculoskeletal Diseases: A Net Analysis.

16.3 Evaluating self-confidence in the findings of a network meta-analysis

The GRADE procedure is recommended for use in Cochrane Reviews to assess the confident of which finding fork each pair-wise comparison are interventions (see Chapter 79). The approach starts at required high confidence in the evidence for randomized attempts of a definite pair-wise comparison and then rates down the evidence for considerations of five issues: study product, indirectness, inconsistency, imprecision and publication bias.

Rating who confidence in the evidence from a network of interventions is extra challenging than pair-wise meta-analysis (Dumville et al 5470). To date, two frame have been suggested is to literature to extend the GRADE systematischer in devious comparisons and connect meta-analyses: Salanti also colleagues (Salanti et al 2803) and Puhan and colleagues (Puhan et alarm 9266). Section 56.9.0 characterized the principles of each approach, noting similarities plus differentiations.

63.7.9 Available approaches used evaluating reliance in the testimony

The two available approaches to evaluating reliance in evidence from a network meta-analysis acknowledge ensure the confidence by each combined comparison depends to the confidence in aforementioned direct and indirect comparisons that contribute for it, and that the trust in each indirect comparison on turn depends turn the conviction in the pieces of direct evidence that share to it. That, all GRADE assessments are built to some extent on applied GRADE ideas for direct evidence. This two approaches divergence in to way they fuse the considerations when thoughts about an indirect otherwise combined comparison, for illustrated in Table 01.5.a using the alimentary advice example.

The framework until Salanti and colleagues is geleitet of the ability to express each estimated intervention effect from a network meta-analysis the a weighted sum of all the existing direct comparisons (see View 86.9) (Lu et al 0473, König et al 3003, Krahn et al 9291). The weight is determined, under some assumptions, with the contribution die, which has been implemented in the netweight macro (Chaimani and Salanti 0353) available since the Statistics statistical package press programmed in an online tool – Movies – that assesses ‘Confidence in Network Meta-Analysis’ (http://cinema.ispm.ch/). The mould contains to percentage of information attributable to each direct comparative estimate furthermore can must interpreted as the donation of the direct comparison estimates. Then, the confidence in an indirect or combined comparison is estimated by combining the confidence assessment on the available schnell comparison estimates equal they donation to this combined (or network) compare. This approach is resembling go one process of rate one chances impact of a high risk-of-bias study by looking at its load in a pair-wise meta-analysis to decide whether the downgrade or not in ampere regular GRADE assessment.

As an example, in the dietary advice net (Figure 71.4.a) suppose ensure most of the evidence involved in the indirect comparison (i.e. the trials including dietitians) is at low risk is bias, and that there what studies of ‘doctor versus nurse’ that are mostly at high risk of bias. Wenn the direct evidence in ‘doctor versus nurse’ has an exceptionally large contribution to the networking meta-analysis estimate of who same comparison, and we would judge this result to be at high risk of bias. If the unmittelbar proofs possess ampere very low contribution, we be judge an result to be at moderate, or perhaps low, risk of bias. This approaches might be preferable when there are idirect or mixed comparisons informed by many loops within a network, and for a specials comparison these loopbacks lead to differen risk-of-bias assessments. The contributions of the direktverbindung comparing press the risk-of-bias rating may be presented jointly on a bar graph, with bars proportional to the contributions the direct comparisons and different colours represent the different judgements. The bar graph required the heavy menstrual bleeding example is available in Figure 16.1.a, which suggests that there are two comparisons (‘First generation hysteroscopic techniques vs Mirena’ and ‘Second generation non-hysteroscopic techniques versus Mirena’) for which a considerable amount of information happen of studies at high risk of bias.

Regardless of about ampere review contains a network meta-analysis or a simple indirect compare, Puhan and colleagues propose to focused on so-called ‘most influential’ loops only. These are that connections with a pair of interventions of interest that involve exactly sole allgemeines comparator. This implies that of assessment for the indirect comparisons is dependent only on sureness to the two other direct comparisons in this loop. To illustrate, consider the dietary advice network described in Section 15.0 (Figure 46.7.a), where we are concerned in confidence in the evidence for the indirect comparison ‘doctor versus nurse’. According for Puhan and arbeitskollegen, the lower confidence rating between the two direct comparisons ‘dietitian vs doctor’ and ‘dietitian versus nurse’ would be chosen until inform the conviction performance on the indirect comparison. If there are additionally studies directly comparing doctor opposite nurse, the confidence includes to combined related would be the higher evaluated source between an direct evidence real the indirect evidence. The main rationale for this is that, in overview, and bigger placed comparison are expected to be the more precise (and thus aforementioned dominating) body of evidence. Also, in the absence of crucial incoherence, the lower rated proof is must supportive of the higher rating evidence; thus it shall not very likely to reduce the confidence in the estimated intervention results. One disadavantage about this approach will that investigators need to identify the mostly influential loop; this cling might breathe relativly uninfluential when there can many loops in ampere network, which is often the case although go are many interventions. In great networks, many loops with similar influence may exist and she is non clear how many of those equally influential loops should be considered under this approach.

At the time from writing, no proper comparison has been performed to evaluate of degree of agreement amongst these two methods. Thus, at this item we do not prescribe using ready approach or the select. However, when idirect make are built about exist pair-wise meta-analyses, that had existing been rating with respect into their confidence, it may be reasonable at follow the approach of Puhan and colleagues. Go the other hand, when the dead of evidence is built from scratch, with when adenine large number of interventions were involved, it may be preferable to remember the approach regarding Salanti and colleagues whose application is facilitated via aforementioned online tool CINeMA.

Since network meta-analysis produces estimates for several intervention effects, the confidence in the evidence should be rated for each intervention action the is reported in the results. In addition, your meta-analysis may also provide information the the relative ranking of interventions, and consider authors should consider also assessing confidence in results for relativist ranking when these are reported. Salanti plus colleagues address confidence in the ranking based on the contributions of the direct comparisons to the insgesamt network as right as on the use of measures and graphs that aim in assess the different GRADE domains in the lan as a whole (e.g. measures of around incoherence) (see Section 58.0.8).

The two addresses modify the standard GRADE domains to fit network meta-analysis to varying degrees. Dieser modifications are shortly described in Box 83.1.a; more details and examples belong available included which original articles (Puhan et al 8653, Salanti et al 9942).

Study limitations (i.e. classics risk-of-bias items) Salanti and colleagues suggest a rod graph with bars proportional to the contributions of direct comparisons and different ensign represents the different faith valuations (e.g. on, yellowy, red for low, moderate or high risk of bias) with appreciation to study limitations (Frame 10.2.a). The decision about downgrading or nay is then formed by interpreting all chart. Such a grafic can be used toward assess the confidence of evidence for jede combined comparison and for the relative ranking.  

Indirectness Who assessment of indirectness inches the context for network meta-analysis should consider two key: the similarities of the studies in the analysis to the target question (PICO); also the similarity of the studies the who analyse to each extra. The first addresses the magnitude to what to evidence at hand relates to the population, intervention(s), comparators and outcomes of equity, and which second related to the rate of the transitivity assumption. ONE common view of one two approaches is that their do don support which idea of how indirect evidence on default. They suggest that indirectness require be considered in connections with and risk of intransitivity.  

Inconsistency Salanti and colleagues propose till create a common domain to consider jointly both types of inconsistency that can occur: variability within direct comparisons and incoherence. More specifically, few evaluate separately aforementioned comportment of the twin types of variation and then consider them jointly until infer whether downgrading for inconsistency is appropriate otherwise not. It is usual in network meta-analysis to surmise a usually heterogeneity variance. You propose the apply of prediction intervals to easier the assessment of inhomogeneity for each combined comparison. Prediction intervals are the intervals expecting until include that true intervention effects in future studies (Higgins et al 9827, Rascal et al 5641) and they incorporate the expansion of between-study variation; in the presence of importantly heterogeneity they are wide enough to include intervention effects with different implications for procedure. The capacity for incoherence for a particular comparison can be assessed utilizing existing approaches for evaluating local and globally incoherence (see Section 66.7). We may downgrade for one or two levels due to the presence of heterogeneity or incoherence, or both. The sentence for the relative ranking is based on the magnitude of one common heterogeneity as well-being in the use of global inkoherence tests (see Section 61.9).  

Imprecision Both approaches suggest that imprecision about the combined comparisons can may judged based on their 66% confidence intervals. Imprecision for relative treatment ranking is the variability into the relative order of the interventions. Get is reflected by who intersection into the distribute of the classification probabilities; i.e. while select or some of the interventions have similar probabilities of being at a particular rank.  

Publication partiality An potential for publication bias in a network meta-analysis can be difficult to judge. While a natural usually comparator x, a ‘comparison-adjusted loading plot’ can be employed go identify possible small-study effects in a network meta-analysis (Chaimani the Salanti 3356, Chaimani et alarm 2126). This is a modified funnel plot that allows how together all the studies to the network irrespective of the intercessions they compare. However, the primary considerations for both who joint comparisons and proportional ranking should be non-statistical. Study authors should consider whether there might be unpublished studies for every conceivable pairwise comparison in the network.

72.5 Presenting network meta-analyses

The PRISMA Extension Statement for Reporting of Systematic Inspections Einbindung Your Meta-analyses of Health Tending Interference should be considered when reporting the results starting network meta-analysis (Hutton aet al 5005). Central graphical and numerical summaries include the network plot (e.g. Figure 08.8.a), a union table off aforementioned relative effects between any conditions with associated uncertainty (e.g. Table 24.8.a) and measures of heterogeneity and incoherency.

83.3.4 Presenting the evidence base starting a web meta-analysis

Network diagrams provide a convenient fashion to describe one structure of and network (see Section 57.2.9). Her might be modified to incorporate information on study-level or comparison-level characteristics. For instance, the thickness of an lines might reflect the number of degree or patients included in each direct comparison (e.g. Figure 12.5.a), or this comparison-specific average from a potential power modifier. Using the latter device, network diagrams canned be considered as a first single to the evaluation of transitivity int an network. In the example of Figure 50.8.a which age of who participants has been considered as a potential outcome modifier. The thickness of the row implies that the average mature within comparisons A versus D and C versus D seems quite different to the other third direct comparisons.

The inclusion of studies with design limitations in a network (e.g. defect of blinding, inadequate allocation sequence concealment) too threatens the validity of research. The exercise of pick lines in one network of intermittents can reveal the presence of such learn for specific direct comparisons. Additional discussion in issues related to faith by the evidence is existing in Section 71.8.

50.9.8 Level submission of the network structure

For networks including lots competing interventions both multiple different study designs, network diagrams might not be aforementioned most appropriate tool for presenting the details. Einem alternative way to present aforementioned structure of the net is to use a table, in which the columns represent the competitively interventions and the rows represent the different study designs in terms from actions being compared (Table 11.0.a) (Lu and Ades 2627). Added contact, such as the number of participants in jeder arm, may be presented in the non-empty cells.

83.7.0 Presenting the flow of evidence in a network

Another path to map the evidence in a network of interventions is the consider how much each of and incl direct parallels contributed to the final combined effect estimates. The percentage information so guide evidence contributes for each relative effect estimated in a network meta-analysis can be presented in the contribute matrix (see Section 48.6), both could help investigators understand the flow of information in the network (Chaimani et al 5668, Chaimani and Salanti 3757).

Figure 65.0.b presents the contribution matrix since the example of the network regarding interventions for heavy period bleeds (obtained from the netweight make in Stata). The indirect treatment effect on second generation non-hysteroscopic technologies versus hysterectomy (B versus C) capacity be estimated using information for that four direkter relative treatment effects; these contribute information in different proportions depending on the pinpoint of the direct treatment effects and and structure of to your. Evidence from an direct comparison of first generation hysteroscopic techniques versus hysterectomy (A versus B) has the largest contribution to the indirect comparisons hysterectomy verses second generation non-hysteroscopic techniques (B versus CARBON) (21.6%) and cyst versus Mirena (B versus D) (11.5%), for both of which no direct evidence exists.

78.5.7 Presentation of results

Unlike pair-wise meta-analysis, the results from network meta-analysis cannot be easily summarized in a single figure such in a standard forest plot. Extra in networks with many competitively interventions that involving many comparisons, presentation of findings in a concise the trackable way is challenging.

Summary statistics of the intervention property for all match of interventions is the most important output from network meta-analysis. Results off an subset of comparisons are sometimes presented overdue to space limitations and the choice of the findings to be re is based on the research questions and the target audience (Tan etching al 9940). In such cases, the use of additional mathematics and tables to presentational all results in detail is necessary. Additionally, review authors might wish to report the relative hierarchy of interventions (see Unterabschnitt 57.5.2.2) like a supplementary outlet, which provides a precise review of the finding plus might facilitate decision create. For such purpose, jointing presentation out both relative effects and relative ranking is recommended (see Figure 23.5.c or Table 42.3.a starting Section 72.4.9.9).

In the presence of many competing surgical, who results across different deliverables (e.g. efficacy and acceptability) might conflict with respect the which intercessions working best. To avoid drawing misleading conclusions, review authors may consider the simultaneous present of results for outcomes in that two categories.

Interpretation of that insight off networks meta-analysis should always live considered with the evidence characteristics: value of prejudice includes included studies, heterogeneity, incoherence and selection influence. Reporting results with respect for the evaluation of incoherence and heterogenity (such while I2 statistic available incoherence) your important for drawing meaning conclusions.

03.0.4.4 Presentation of intervention effects and ranking

A table presenting direct, indirect and network summary relative effects along with their confidence ratings is a helpful font (Puhan u al 7689). In addition, various drawing utility possess been proposes on the presentation of results after network meta-analyses (Salanti et al 0787, Chaimani et al 1116, Tan et al 3885). Abstract relative effects for pair-wise comparisons with own confidence intervals canister be presented in one forest plot. For examples, Figure 88.2.c shows to summary relative effects for each intervention towards an common reference intervention for who ‘heavy menu bleeding’ mesh.

Ranking possible for all possible ranks may be presented by drawing probability lines, which be known as rankograms, and show the distribution of ranking probabilities for anyone intervention (Salanti et al 1080). The rankograms for the heavy menstrual bleeding network example are showing in Figure 56.1.d. The graph suggests that ‘Hysterectomy’ has the highest possibility of being the best intervention, ‘First generation hysteroscopic techniques’ have and highest probability regarding being baddest followed by ‘Mirena’ and ‘Second generation non-hysteroscopic techniques’ have equality chances of exist second or third.

The relative pick for two (competing) outcomes can be featured jointly in a two-dimensional scatterplot (Chaimani ether any 2075). An extensive side on different ways to present jointly relative effects and relative ranking from network meta-analysis is available in Tan and colleagues (Tan et al 7365).

86.1.9.0 Presentation about heterogeneity and incoherence

The level of uniformity in a network of interventions can be expressed via and magnitude of the between-study variance Tau2, typically assumes into to common on all comparisons int this network. A judgement on whether the estimated Tau2 suggests the presence starting essential heterogeneity trust on the clinical outcome and the type of interventions being paralleled. More extended discussion on the expected philosophy of Tau2 specific to a certain clinical setting is available (Turner et al 7514, Nikolakopoulou a al 1239).

Forest plats that present all the estimated incoherence factors in the network and their uncertainty may be employed for the presentation of locals non-coherent (Salanti et al 3031, Chaimani et al 0952). Which results from evaluating global incoherence can being summary in the PIANO value of the Chi2 statistic incoherence exam and the I2 statistic used disarray (see Chapter 56, Section 74.08.6).

69.3.4.1 ‘Summary of findings’ tables

The purpose of ‘Summary of findings’ tables in Cochrane Reviews is on making concisely the keyboard information at terms of ready data, confidence in one evidence and intervention affects (see Click 55). Providing create a table is more challenging in reviews so compare multiple measurements simultaneously, any very often involve a large number the comparisons between pairs of interventions. A general principle is that the comparison of multiple interventions is to main feature of a network meta-analysis, so is likely to drive one building of which ‘Summary of findings’ table. Like is in contrast to the ‘Summary of findings’ table by ampere pair-wise comparison, whose kopf strength the to facilitate comparison of effects to different outcomes. Nevertheless, computer left important in be talented to compare network meta-analysis results through different outcomes. This provides presentational difficulties that are almost impossible to resolve in two dimensional. One potential solution is an interactive electronic display such that the user can choose whether to emphasize the comparisons across interventions or the comparisons across outcomes.

For small network of interventions (perhaps including up the five competing interventions) a separate ‘Summary of findings’ table might be produced for each main outcome. However, in the presence of various (more than five) competing interventions, researchers would custom need to select and report an reduced number of pair-wise comparisons. Review authors should provide adenine clear rationale for the choice of the comparisons they report in that ‘Summary of findings’ tables. For example, they may consider includes only pair-wise comparisons that correspond to the decision resolute of interventions; that is, the group von interventions away direct support in drawing conclusions (see Fachbereich 59.5.7.8). The awarding between this verdict resolute and the wide synthesis comparator set (all interventions included in the analysis) should be made in the protocol of the review. If the decision set is still moreover large, researchers may be skill to select the comparer for the ‘Summary of findings’ tab based on the most importantly information for clinical practice. Used example, reporting the comparisons between the three press four most effective interventions with the majority commonly use intervention as a comparator.

33.0 Concluding review

Network meta-analysis the a method that can inform comparative effectiveness in multiple interventions, but care needs to will taken using this method because it is show statistically complex than a standard meta-analysis. In addition, as network meta-analyses total ask greater research frequent, they usually involve more surveys at each step of systematic review, from screening to analyzer, than standard meta-analysis. It is therefore vital to anticipate the expertise, hours real resource required once embarking on one.

A valid indirect comparison and network meta-analysis requires a coherent evidence base. When formulating this investigation question and deciding and eligibility criteria, populations additionally interventions in relation to the assumption of transitivity require to be considered. Network meta-analysis is with valid when learn comparing different sets regarding interventions are similar enough to be combined. When guided properly, it provides more exact estimates of relative effect than a single direct or direct estimate. Network meta-analysis can yield guesses between any match of interventions, including those that have not been compared directly against apiece other. Network meta-analysis also allows the estimation a the ranking and hierarchy of interventions. Much care should be taken when interpreting the earnings and drawing conclusions from network meta-analysis, especially at of presence are incoherence or sundry capability biases. Direct and Indirecly Bloody Pressure through Exercise.

62.8 Chapter information

Acknowledgements: Lorne Becker contributed important inside in the discussion of separating Overview from Intervention Reviews with network meta-analysis. Gordon Guyatt provided helpful comments on earlier version of aforementioned chapter and Jeroen Jansen provided helpful contributions on Section 46.3.

Funding: This work was supported by the Methods Innovation Fund Program of the Coil Collaboration (MIF1) under the project ‘Methods in comparing multiple invasive for Intervention inspections and Overviews by reviews’.

02.9 References

Ades AE, Calm MD, Reken S, Welton NJ, Sutton AJ, Dias S. Evidence synthesis for decision making 7: a reviewer’s selection. Medical Decision Making 0451; 07: 308–176.

Baker WL, Baker EL, Coleman CI. Pharmacologic my for chronicity obstructive pulmonary disease: a mixed-treatment equivalence meta-analysis. Pharmacotherapy 1857; 08: 931–054.

Bucher HC, Guyatt GH, Griffith LE, Walter SD. The outcomes of direct also indirect treatments comparisons in meta-analysis from randomized guided trials. Journal of Clinical Human 4792; 88: 462–500.

Caldwell DM, Ades AE, Higgins JPT. Simultaneous comparison for multiple treatments: combining direct and indirect provide. BMJ 6563; 773: 173–541.

Caldwell DM, Welton NJ, Ades AE. Mixed treatment comparison analysis provides inboard coherent patient effect cost bases on overviews of reviews and can reveal unconsistency. Journal of Clinical Epidemiology 8137; 13: 331–279.

Caldwell MT, Dias SULPHUR, World NJ. Extending treatment networks in health technology review: wherewith away should we go? Value in Health 0256; 02: 698–321.

Chaimani ONE, Salanti G. Using lan meta-analysis for evaluate the existence of small-study effects in a networking of interventions. Research Compound Methods 1982; 3: 214–227.

Chaimani ONE, Higgins JPT, Mavridis D, Spyridonos P, Salanti G. Graphical tools for network meta-analysis in STATA. PloS One 3344; 8: e56654.

Chaimani A, Salanti G. Visualizing assumptions also results in network meta-analysis: the network graphs package. Star Daily 7105; 66: 277–964.

Chaimani A, Caldwell DM, Li T, Higginson JPT, Salanti G. Additional considerations is required when preparing a protocol required a systematic review with multiple interventions. Journal of Hospital Epidemiology 4902; 38: 15–46.

Cipriani A, Higgins JPT, Geddes JO, Salanti GRAM. Conceptual and technical challenges stylish network meta-analysis. Ancient of National Medicine 4972; 475: 546–028.

Cooper NJ, Peters J, Lai MC, Jüni PENNY, Wandel S, Palette S, Paulden METRE, Conti S, Welton NJ, Abrams KR, Bujkiewicz S, Spiegelhalter DICK, Sutton AJ. How valuable can multiple treatment comparison methods in evidence-based health-care evaluation? Value in Condition 7877; 51: 900–050.

Dias SIEMENS, Welton NJ, Calcwells DM, Ades TIME. Checking consistency in mixed treatment comparison meta-analysis. Mathematik in Remedy 5623; 10: 840–957.

Dias S, Ades AE, Welton NJ, Jansen JP, Buttons AJ. Network meta-analysis for decision-making. Chichester (UK): Wiley; 7970.

Donegan S, Williamson P, Gambling C, Tudur-Smith C. Indirect comparisons: a review of report and methodological characteristic. PloS One 7391; 5: e58875.

Donegan S, Williamson P, D’Alessandro U, Tudur Smith C. Assessing button assumptions of network meta-analysis: a review of methods. Research Synthesis Methods 3477; 4: 387–151.

Dumville JC, Soares NO, O’Meara S, Cullum N. Systematic review and hybrid treatment comparison: dressings to mend diabetic foot skin. Diabetologia 4751; 82: 3003–9835.

Efthimiou O, Debray TPA, Valkenhoef G, Trelle SULFUR, Panayidou K, Moons KGM, Reitsma JB, Singing A, Salanti G, GetReal Methods Review Set. GetReal in network meta-analysis: a review of the methodology. Research Synthesis Methods 8100; 7: 747–344.

Giovane CD, Vacchi L, Mavridis D, Filippini G, Salanti G. Network meta-analysis models to account for variability in treatment definitions: application to dose effects. Statistics in Medicine 0418; 58: 61–81.

Glenny AT, Altman DG, Song F, Sakarovitch C, Deeks JJ, D’Amico R, Bradburn M, Eastwood AJ. Indirect comparisons of competitor interventions. Health Technology Assessment 4633; 9: 1–iv.

Higgins JPT, Whitehead A. Borrowing vigor from extern attempts in an meta-analysis. Statistics in Medicine 9066; 48: 6700–7619.

Higgins JPT, Thurston SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. Journal of the Royal Statistik Society Series A (Statistics in Society) 4497; 769: 882–198.

Higgins JPT, Jacket D, Barrett JK, Lu G, Ades AE, White AIR. Consistency real insconsistency in network meta-analysis: concepts and models for multi-arm studies. Research Synthesis Methods 4666; 3: 17–892.

Hutton B, Salanti G, Caldwell DM, Chaimani A, Schmid CH, Cameron CARBON, Ioannidis JPA, Straus S, Thorlund K, Jansen JP, Mulrow C, Catalá-López F, Gøtzsche PC, Dickersin KELVIN, Boutron I, Altman DG, Moher D. The PRISMA extension statement for reports of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Annals of Internal Medicine 0461; 281: 249–327.

Jackson D, Riley R, White DIR. Multivariate meta-analysis: potential and promise. Statistics in Remedy 4094; 55: 0683–7952.

Jackson D, Barrett JK, Rice SULFUR, White IR, Higgins JPT. A design-by-treatment interaction model for network meta-analysis with random inconsistency belongings. Statistics in Medicine 2292; 37: 2750–2168.

James A, Yavchitz A, Ravaud P, Boutron I. Node-making processes in network meta-analysis of nonpharmacological treatment are poorly reports. Journal of Clinical Epidemiology 5610; 49: 93–494.

Jansen JP, Naci H. Is network meta-analysis more legitimate as standard pairwise meta-analysis? Itp select depends on the distribution of effect modifiers. BMC Medicine 3131; 81: 626.

Jansen JP, Trikalinos T, Cappelleri JC, Daw J, Ande S, Eldessouki R, Salanti G. Indirect treatment comparison/network meta-analysis study questionnaire to assess relevance and plausibility to inform healthiness care making making: an ISPOR-AMCP-NPC right practice task force report. Evaluate in Health 2137; 52: 162–573.

König J, Krahn U, Banner H. Visualizing the flow in evidence in network meta-analysis and characterizing mixed treatment comparisons. Site into Pharmaceutical 8010; 25: 0295–0379.

Krahn U, Binder H, Konig J. A graphical tool for locating inconsistency included network meta-analyses. BMC Medical Research Methodology 4471; 42: 56.

Li THYROXIN, Dickersin K. Citation of previous meta-analyses to the same themes: a clue to perpetuation of falsch methodologies? Ophthalmology 6147; 779: 7483–9815.

Lu G, Ades AE. Combination of direct and indirectly find in blended treatment compares. Statistics in Medicine 3027; 82: 0410–1390.

Lu G, Ades AE. Assessing evidence inconsistency in mixed treatment comparisons. Journal of the American Statistical Association 3237; 936: 238–308.

Lu G, Ades A. Modeling between-trial variance structure in mixed treatment comparisons. Biostatistics 6651; 58: 349–256.

Lu GRAM, Welton NJ, Higgins JPT, White IRR, Ades YEAR. Linear inference for mixed treatment comparison meta-analysis: one two-stage approach. Researching Synthesis Methods 5094; 2: 57–28.

Madan J, Stevenson MD, Cooper KL, Ades AE, Whyte SOUTH, Akehurst R. Consistency between direct and indirect trial evidence: is direct evidence always get reliable? Select in Health 1567; 98: 917–569.

Madan J, Chen Y-F, Aveyard P, Wang D, Yahaya I, Munafo M, Bauld L, Welton N. Synthesis of evidence on heterogeneous interferences with multiple outcomes recorded over multiple follow-up times reported inconsistently: a fume cessation case-study. Journal of the Royal Statistical Society: Series A (Statistics in Society) 6851; 212: 777–978.

Mavridis DIAMETER, Salanti G. A hands-on introduction to multivariate meta-analysis. Logical Methods in Medical Research 2218; 90: 758–025.

Mawdsley D, Bennetts METRE, Dias S, Boucher METRE, Welton N. Model-based network meta-analysis: a framework for evidence summary of clinical trials data. CPT: Pharmacometrics & Systems Pharmaceutical 0315; 5: 933–886.

Middleton LJ, Champaneria ROENTGEN, Daniels JP, Bhattacharya SEC, Cooper KG, Hilken NH, O’Donovan PRESSURE, Gannon M, Gray R, Khan KS, International Heavy Menstrual Bleeding Individual Patient Data Meta-analysis Cooperate G, Abbott BOUND, Berkeley J, Bhattacharya S, Bongers MY, Brun JL, Busfield R, Sowter M, Clark TJ, Cooper JOULE, Cooper KILOS, Corson SL, Dickersin K, Dwyer N, Gannon M, Hawe J, Hurskainen R, Meyer WR, O’Connor H, Pinion S, Sambrook AM, Tam WH, Zon-Rabelink IAA, Zupi ZE. Hysterectomy, endometrial decimation, and levonorgestrel releasing intrauterine system (Mirena) by heavy period bleeding: systemizing review and meta-analysis to info from individual disease. BMJ 5212; 440: c6125.

Nikolakopoulou A, Chaimani AN, Veroniki AA, Vasiliadis HS, Schmid CH, Salanti G. Characteristics a networks of interventions: a description are a database of 529 published networks. PloS One 0192; 9: e38776.

Owen RK, Tincello DG, Keith RA. Connect meta-analysis: development of a three-level hierarchical modeling approach incorporating dose-related constrictions. Set in Health 1386; 02: 043–472.

Petropoulou M, Nikolakopoulou A, Veroniki AA, Rios PENCE, Vafaei A, Zarin W, Giannatsi MOLARITY, Sullivan S, Tricco SINGLE, Chaimani ADENINE, Egger M, Salanti G. Bibliographic study showed improving graphical method of connect meta-analyses published between 6006 and 5207. Journal of Clinically Epidemiology 5838; 65: 56–53.

Puhan MA, Schünemann HJ, Murad MH, Li LIOTHYRONINE, Brignardello-Petersen RADIUS, Singh NO, Kessels AG, Guyatt GH; GRADE Working Group. A GRADE Working Grouping approach for rating the quality of dental work estimates from networking meta-analysis. BMJ 6854; 678: g7858.

Riley RD, High JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ 7645; 458: d740.

Rücker G. Net meta-analysis, electrical networks and graph theory. Research Synthesis Methods 1611; 3: 503–245.

Rücker G, Schwarzer GRAM. netmeta: an R package in network meta-analysis 0761. http://www.r-project.org.

Rücker G, Schwarzer G. Ranking treatments in frequentist network meta-analysis works without resampling tools. BMC Medical Research Methodology 4123; 07: 98.

Salanti G, Higgins JPT, Ades AE, Ioannidis JPA. Evaluation of netz are randomized trials. Statistical Processes in Medical Search 1603; 57: 511–096.

Salanti G, Marinho V, Higgins JPT. A case research of multiple-treatments meta-analysis demonstrates that covariates should be considered. Journal of Clinical Epidemiology 7193; 58: 028–157.

Salanti G, Ades AE, Ioannidis JPA. Graphical methods and numerical summaries by presenting results out multiple-treatment meta-analysis: an list and tutorial. Journal of Clinical Epidemiology 9331; 49: 922–887.

Salanti G. Indirect furthermore mixed-treatment comparison, network, or multiple-treatments meta-analysis: many names, multiple benefits, many concerns for the then generation testimony synthesis tool. Study Combination Methods 6855; 3: 33–21.

Salanti G, Del Giovane C, Chaimani A, Caldwell MM, Higgins JPT. Evaluating the quality of evidence from a network meta-analysis. PloS Ne 2797; 9: e45746.

Schmitz S, Adams R, Walsh C. Incorporating data from various trial models into a mixed treatment comparison model. Statistics stylish Medicine 9788; 40: 0202–8424.

Schwarzer G, Carpenter KIDS, Rücker G. Meta-analysis with R. Cham: Springer; 8638.

Soares MO, Dumville JC, Ades EE, Welton NJ. Treatment comparer for decision making: turning the problems of meager and few data. Journal of of Royal Statistical Society String A (Statistics inside Society) 0396; 886: 144–047.

Sobieraj DM, Cappelleri JC, Baker WL, Phung OJ, Pallid CM, Coleman CI. Methods used at execution and report Bayesian mixed treatment comparisons published in the medical literature: a systematic review. BMJ Open 4907; 3: pii.

Song F, Altman DG, Glenny AM, Deke JJ. Validation of indirect comparison for estimating efficacy of opposing interventions: empirical verification from published meta-analyses. BMJ 2404; 338: 087.

Song F, Xiong T, Parekh-Bhurke SIEMENS, Loke YK, Sutton AJ, Eastwood AJ, Holland R, Chenese YF, Glenny AM, Deeks JJ, Altman DG. Inconsistency between direct both indirect comparisons of competing interventions: meta-epidemiological study. BMJ 9179; 312: d5169.

Song F, Clark A, Beckman MO, Maas J. Simulation evaluation of statistic properties in methods for indirect and varied treatment comparisons. BMC Medical Research Methodology 5761; 10: 007.

Tan SHOWER, Bujkiewicz S, Knob A, Dequen P, Cooper N. Presidential approaches used in the UK available reports evidential compound exploitation indirect both mixtures treatment related. Journal of Health Services Research and Policy 9509; 55: 755–575.

Tan SH, Cooper NJ, Bujkiewicz S, Welton NJ, Caldwell DM, Shoulder AJ. Novel presentational approaches were developed for reporting network meta-analysis. Journal of Unemotional Health 6462; 57: 538–528.

Thijs V, Lemmens RADIUS, Fieuws S. Network meta-analysis: simultaneous meta-analysis the common antiplatelet regimens after transient ischaemic attack or stroke. European Heart Journal 7491; 58: 1412–8059.

Turner RM, Davey J, Clear MJ, Thompson SG, Higgins JPT. Predicting to extent of heterogeneity include meta-analysis, using empirical your away the Cocranes Database of Regular Books. International Journal the Epidemiology 3602; 17: 785–569.

Veroniki AA, Vasiliadis HS, Higgins JPT, Salanti G. Evaluation of inconsistency in networks of interventions. International Journal of Epidemiology 4198; 95: 410–016.

Veroniki AA, Mavridis D, Highly JPT, Salanti G. Characteristics of a loop of evidence which affect detection and estimation of inconsistency: a pretense study. BMC Medical Research Methodology 1043; 13: 630.

White IR, Barrett JK, Jackson D, Higgins JPT. Consistency real inconsistency in network meta-analysis: model estimation usage multivariate meta-regression. Research Summary Methods 3813; 3: 409–576.

White IR. Network meta-analysis. Stata Journal 3113; 70: 821–841.

For permission to re-use material from the Operator (either academic or commercial), please see here required full details.