© 2006 1 literature review designs narrative review systematic review meta-analysis

1© 2006

Literature review designs

Narrative ReviewSystematic Review

Meta-analysis

Evidence-based Chiropractic © 2006-72

Literature review

• Defined as a systematic, explicit, and reproducible way of identifying, evaluating, and interpreting all of the research findings and scholarly work available on a topic – A high-quality review is not haphazard– Ideally, all of the existing work should be

included

• Considered descriptive or observational


Meta-analysis Quantitatively combines the results of studies that are the result of a systematic literature review. Capable of performing a statistical analysis of the pooled results of relevant studies.

The three types of literature review designs

Narrative Review Selective review of the literature that broadly covers a specific topic. Does not follow strict systematic methods to locate and synthesize articles.

Systematic Review Utilizes exacting search strategies to make certain that the maximum extent of relevant research has been considered. Original articles are methodologically appraised and synthesized.


Narrative reviews

• Summarize in general what is in the literature on a given topic– Often written by experts in a given field– A good source for background information

• Do not follow strict systematic methods like the other literature review designs – Therefore, they are prone to bias– Lower in the hierarchy of evidence


Narrative reviews (cont.)

• Authors like to write them because they are relatively easy to carry out and compose

• Practitioners like to read them because they are easier to comprehend than more complex designs – Do not require a lot of background knowledge

to understand the message


Narrative reviews are prone to bias

• They do not employ many of the safeguards needed to control against bias – Authors may be selective as to which articles

are included – They may include articles that support their

hypothesis and exclude those that do not – Rigorous appraisal methods are not used to

evaluate included articles


Prone to bias (cont.)

• During the literature search– Authors have their own opinions on the topic

and may try to find studies that support their viewpoint and overlook conflicting studies

• During the synthesis of the literature – The approach to analyzing the collected

information is often subjective and disorganized


Prone to bias (cont.)

• In the discussion and conclusion– The authors’ opinions may be mixed together

with evidence – Authors may simply count the number of

studies on each side of an issue and then espouse the view presented by the majority without considering the strength of each study

• e.g., quality, research design used, the effect size, and sample size


Selection bias innarrative reviews

• a.k.a., reference bias

• Occurs when authors choose articles that support their own conclusions and exclude articles with conflicting views – Results in an erroneous representation of the

literature – The review may lead undiscerning readers

astray


Systematic reviews

• Use strict methods to locate, appraise and synthesize all research on a topic– Similar to narrative reviews, but with improved

procedural quality

• Designed to answer specific clinical questions where several primary studies exist– Thus, a good source of clinical evidence


Systematic reviews (cont.)

• Articles are evaluated using appraisal instruments – In an attempt to achieve impartiality – More reproducible

• Dissimilarities between the findings of studies are investigated

• Multiple reviewers are usually involved– Any disagreements are resolved



• The search and selection criteria for articles are well defined

– Multiple databases should be searched – Explicit inclusion criteria

• The results of the included studies are qualitatively or quantitatively synthesized

– Qualitative – written information is merged– Quantitative – data are merged



• Because of strict methodology and thoroughness, conclusions are typically less biased than narrative reviews

• Nevertheless, they may still be influenced by the authors’ opinions

– Still a potential for selection bias– Criteria may be applied differently when

appraising included studies– Therefore, must be critically appraised



• No widely accepted method exists for assessing the validity of studies

– i.e., the process is a judgment call – Thus, reviewers sometimes disagree

• Information derived from a systematic review may be too narrow to completely answer a specific clinical question

– Given that they are designed to answer focused questions about patient care


Differences between narrative and systematic reviews

Feature Narrative Review Systematic Review

Topic Typically broad-scoped Focused research question

Data sources and search strategy

The search strategy and databases that were used may not be provided

The search strategy is explicit and comprehensive with a list of all databases that were utilized

Authorship A recognized expert(s) on the topic

A team of experts having methodologic and clinical expertise

Article selection criteria

Typically not specified Consistently applied inclusion and exclusion criteria

Searching May be extensive, intended to locate literature on the topic area in question

Extensive, intended to locate all primary studies on a particular research question

Appraisal of included articles

Indefinite, may be variable Critical appraisal is meticulous, typically involving the use of data extraction forms

Synthesis A qualitative summary is usually provided

A qualitative summary is provided, quantitative when the data can be pooled

Inferences Sometimes evidence-based Usually evidence-based


Systematic reviews typically only include RCTs

• Studies are included primarily based on their quality – Most systematic reviews only include RCTs

because it is the only design that adequately controls for confounding variables and biases

– The potential for studies to overrate the treatment effect is higher when bias is present

• Studies using other less rigorous designs are usually eliminated


Conducting systematic reviews


Search strategy

• Should be described in enough detail so that another researcher could replicate the results, including: – Database(s) searched– Date the search was performed– Time-frame encompassed by the search– A list of search terms used– Languages


Search strategy (cont.)

• Conference proceedings, unpublished studies, and hand-searching of journals are sometimes included in the search – If so, the procedures involved should be

described– Authors must justify using unpublished work


Weighting of studies

• Articles may be rejected in a systematic review due to their poor quality

• Alternatively, studies are assigned weights in relation to their assessed validity– Studies that are more valid will have more

influence on the review’s final results– Based on methodological quality, width of the

confidence intervals, and external validity


Publication bias

• Studies with statistically significant results are more likely to get published than those with non-significant results

• Causes of publication bias: – The author or funding source does not

consider a “failed study” worthy of submission– Journals are less likely to publish studies that

fail to show positive results


Publication bias (cont.)

• Reviews affected by this bias tend to give an overoptimistic view of the effectiveness of the therapy– The chance of this bias occurring is reduced

when authors of systematic reviews search sources other than journals

• Publication bias in situ – A type of bias where a portion of a study’s

results are suppressed


Meta-analysis

• a.k.a., quantitative systematic review• A type of systematic review that

statistically combines the results from a number of studies

• Capable of producing a single estimate of the effect of a treatment– Represents the “average” treatment effect– An estimate of the true treatment effect size


Meta-analysis (cont.)

• The same explicit methods as systematic reviews are utilized

• Systematic reviews and meta-analyses are are at the top of the hierarchy of evidence because of their strict methodology


Weighted average

• Meta-analyses typically produce a weighted average for the treatment effect estimate – Small samples are more susceptible to

chance variations than larger studies – Thus, they are given less weight than larger

studies so they will have less influence on the final estimate


Weighting (cont.)

• Weighting is also based on study quality– The quality of the individual studies is rated

and resulting numeric scores are calculated– A corresponding weight is assigned for each

study prior to analysis


Meta-analyses can increase power

• Data from individual studies are combined, which in effect increases sample size – Chiropractic studies commonly involve too few

subjects to detect true differences between the groups

– Pooling data reduces the potential for type II error

– More likely to detect a treatment effect, if there actually is one


Homogeneity and heterogeneity

• Homogeneity – Similarities of included studies that allow them

to be compared– Homogeneity is preferred in meta-analyses– Achieved by using suitable inclusion criteria

• Heterogeneity – Dissimilarities of studies that hamper or even

prevent a realistic comparison of studies


Factors that contribute to heterogeneity

• Heterogeneity in the study samples – Caused by conflicting inclusion and exclusion

criteria, differences in patients’ baseline health status, dissimilar geographical locations of groups, etc.

• Heterogeneity in the study design– E.g., the way dropouts were managed in the

statistical analysis or the length of time allowed for patient follow-up


Factors that contribute to heterogeneity (cont.)

• The way patients were handled – Regarding comorbid conditions, handling of

complications, the control practitioners had in patient care, or the outcome measures used

• Statistical heterogeneity– When the observed treatment effects of

studies are more dissimilar than what would be expected by chance


Consequences of heterogeneity

• When the results of studies in a meta-analysis are inconsistent, it reduces confidence in its conclusions

• The meta-analysis may actually be worthless if too dissimilar – For instance, combining studies that used

different types of comparison groups – Or outcomes that were dissimilar


Forest plot

• A type of graph often used in meta-analyses to illustrate the treatment effect sizes of the studies

• Each study is represented by a black square that is an estimate of their effect sizes

• A horizontal line extends to either side of the squares, the 95% confidence interval


Forest plot


Interpreting a forest plot

• If a study’s 95% CI crosses over the vertical line, it is not statistically significant

• A diamond with a CI line is sometimes presented at the bottom of the forest plot to represent an overall estimate

• The black squares may vary in size representing the weights of the studies


Weighting and overall effect


Effect size

• The difference between the means of the treatment and control groups

• When studies are combined in a meta-analysis, the units of measurement are not always comparable

• Effect sizes are standardized to resolve this problem producing the standardized mean difference


Standardized mean difference

• The effect size divided by the pooled standard deviation – Pooled standard deviation has been adjusted

for the differences in the sizes of the groups

• Represents the standardized difference between group means – i.e., the relative magnitude of the

experimental treatment


Cohen’s d


Odds ratio (OR)

• Cohen’s d is appropriate with continuous data

• An OR is appropriate when the study’s outcome measure is dichotomous– e.g., pain versus no pain

• OR is a comparison of the odds of the outcome being present in the treatment group against the control group


Relative risk (RR)

• a.k.a., risk ratio

• A comparison of the risk of having the outcome in the treatment group with that of the control group

• Sometimes OR is reported and sometimes RR– Experts do not agree on which is most

appropriate


OR vs. RR

• Consider a hypothetical lower back pain study with 25 patients in each group

• 5 in treatment group and 10 in control group are still in pain at the study’s end

OR = = = .38

RR = = = .5

5/20

10/15

.25

.66

5/25

10/25

.2

.4


Meta-analyses are most valid with RCTs

• However, about half of meta-analyses include observational studies – Primarily cohort and case-control

• Observational studies are much more susceptible to biases and confounding than RCTs – Therefore, it is usually inappropriate to

statistically combine the results of such studies


Subgroup analysis

• Meta-analyses typically include patients with a variety of characteristics – e.g., age, gender, condition severity, patient

history, etc.

• Patients in these subgroups may respond to treatment differently– e.g., low back pain patients with leg pain may

respond to treatment differently than low back pain only patients


Subgroup analysis (cont.)

• Carried out to identify variation between patient groups regarding certain outcomes or findings

• The process helps readers to distinguish the effects of a treatment between subgroups

• The statistical power of the subgroups’ will decline as a result


Meta-regression

• A statistical procedure that adjusts for differences between studies in meta-analyses – May be used in subgroup analyses

• Similar to simple regression– Predictor variables: characteristics of the

studies – Outcome variable: treatment effect estimate


Sensitivity analysis

• A type of subgroup analysis that considers non-patient characteristics, e.g., treatment variations or study methodology

• Determines the extent heterogeneity affected the results of a meta-analysis– If the results are weak, sensitivity analysis

may reveal significant treatment effects when different methods are used


Narrative versus systematic reviews and meta-analyses

• There are no strict rules regarding the creation of either type– Therefore, it may be difficult to decide if a

given review is systematic or narrative

• Narrative reviews do not typically use systematic methods– They tend to be subjective and prone to bias– Cover broader topics than systematic reviews


Narrative vs. systematic reviews (cont.)

• May be unclear how conclusions were drawn from the data in narrative reviews – Often the number of studies supporting one

side of a topic is counted and then compared with the number supporting the opposite side

– The side with the highest number of supporting articles wins

– This process does not consider the weight of studies as in systematic reviews


Advantages and disadvantages of narrative versus systematic reviews

Advantages Disadvantages

Narrative reviews

Present a general overview covering a specific topic which provides primary information and/or an updateFairly easy for novice authors to prepare

May not provide the best available answers to focused questionsFindings are less reliable

Systematic reviews

Present a comprehensive review of the literature based on all available research with regard to a focused research questionProvide an estimate of the “true” answer to the research question

Specialized expertise of reviewers is requiredInvolve a formal research protocol Findings are only relevant to a single question


QUOROM Statement

• QUOROM (Quality of Reporting of Meta-analyses)

• The QUOROM Statement was developed to reduce the potential for reviews to reach contradictory conclusions

• An attempt to ensure uniformity and accurate reporting

• Has been adopted by many journals


Structure of review articles

• Should conform to the anatomy of a typical scholarly article – i.e., Abstract, Introduction, Methods, Results,

Discussion, and References

• Literature reviews are in reality a type of research– However, conclusions are derived from

original sources of information


Structured Abstract

• Objective– The author should clearly state the purpose of

the article

• Background – A description of what prompted the review

– Presentation of a context for the review

• Methods– A description of the methods used


Structured Abstract (cont.)

• Discussion – The implication and relevance of the

information the review presents

• Conclusion – Summary of what the review contributes to

the literature

– What new conclusion can be drawn as a result of the synthesis of the literature


Introduction

• Presents the background and context of the problem that inspired review

• The topic should be clearly defined, and tied in with clinical practice

• Novel terms should be defined• A description of the course of the disease,

common outcomes and treatment options• A synopsis of existing research


Introduction (cont.)

• The importance and need for the review should be established by showing – That there are gaps in the literature on the

topic – The extent of the condition’s negative impact

on society in terms of human suffering and monetary costs

– Explained in enough detail to substantiate the need for the review


Introduction (cont.)

• A focused and well-constructed question should be present– Provides direction for the review– Assists readers in determining if the review is

applicable to their individual clinical circumstances

– Should help establish the review’s inclusion criteria


Methods

• Describes the search process and strategies involved, including: – Databases searched – Search terms – Search limits

• e.g., publication years, languages, ages, etc.

• Should include enough detail to enable others to replicate the search


Methods should include

• The criteria that were used to include or exclude studies– For example, exclude surgery related studies

or drug trials

• A description of how studies were appraised– Rating instruments are typically used– However, the reliability of these instruments

varies considerably


Methods should include (cont.)

• Information about– How the relevance of primary studies was

ascertained– How the data were extracted and synthesized– Sources of heterogeneity

• How much• How it was handled


Results

• The outcome of the search process is presented

• Including information on – The number of articles retrieved – How many articles were excluded from the

review and which of the inclusion criteria they failed to meet

• Look for evidence of selective referencing


Results (cont.)

• The Results section is sometimes very short – When short, details about the retrieved

articles are provided in the Discussion section

• However, it may be longer – The characteristics of the included studies

may be described and contrasted in this section

– Often presented in tables


Discussion

• The findings of all of the articles in the review are synthesized to generate a conclusion– There may or may not be a separate

Conclusions section

• Information about the etiology, pathophysiology, diagnosis, treatment, and prognosis of the condition at issue is often provided


Discussion (cont.)

• Presents a new perspective on the topic that is usually more reliable than any of the individual articles in the review

• Caution – authors have the leeway to defend articles that support their viewpoint and challenge those that do not – Systematic methods control for much of this

subjectivity, but it is still possible in the best types of reviews


Discussion (cont.)

• The synthesis is the heart of the literature review design – Consequently it is important to ensure that a

meaningful integration is accomplished

• The author should offer an interpretation of the literature reviewed– A critical appraisal of the articles reviewed

may be in order


Discussion (cont.)

• The conclusion should be in agreement with the evidence presented in the review

• Authors should emphasize what new information can be gained

• The conclusion should not merely repeat what was previously written


References

• Should be comprehensive and cite all articles included in the review

• Derived almost entirely from peer-reviewed journals – But may include conference proceedings,

textbooks, and government documents – Unpublished works too; but keep in mind, they

have not been peer-reviewed


References (cont.)

• No padding of references – Only enough articles should be cited to make

a point

• Be alert for authors who misrepresent sources, which is a fairly common occurrence– e.g., misquote, selectively quote, omit

relevant information, etc.– May have to read source material to detect


Appraisal of review articles

• Recent reviews are typically more useful because they contain the latest research– However, even the latest review may be

behind the most current research

• The literature search should involve multiple databases

• The search strategy should be clearly described, along with the criteria used for study inclusion


Appraisal of reviews (cont.)

• Was a clear study question asked? – Reviews should have a clear purpose in the

form of a study question or statement of purpose

• Especially with systematic reviews, but also applies to narrative reviews

– Otherwise, the review will lack direction and may generate faulty conclusions



• Was the study question focused on a specific clinical issue? – The study question must be focused on a

specific clinical issue that is relevant to patient care to help practitioners make clinical decisions

– Broad topic reviews provide background information, but are not very helpful in answering clinical questions



• Were the inclusion/exclusion criteria that were used to select articles provided and were they appropriate? – Selection criteria should be appropriate for the

clinical question that prompted the search – Difficult to assess the breadth of the review or

judge the extent of the author’s partiality without this information



• Was an adequate literature search conducted? – The search should be capable of finding all

relevant studies – Appropriate databases and search terms – Citations should be collected from the

reference sections of the included articles– How likely is it that relevant studies were

missed during the search?



• Was the validity of the studies included in the review assessed? – Reviews that include articles without

appraising them are poorer quality and less reliable

– Authors may examine issues such as the methods of randomization and blinding, whether concealment of the allocation to groups was used, etc.



• Were the studies appraised in a dependable manner? – Should be reviewed by more than one

reviewer to strengthen confidence in their findings

– Reliable appraisal instruments should be used because they yield more consistent results

• Often included in an appendix



• Were the included studies randomized trials? – Strengthens the usefulness of a literature

review considerably when RCTS are involved– Reviews that include observational studies

provide little evidence to support treatment – However, such reviews are helpful with

patient care and qualify as the best evidence when observational research is all that exists



• Were the results of the included studies similar? – Heterogeneity of studies may occur due to:

• Clinical factors (e.g., dissimilar participants, interventions or outcomes)

• Methodological factors (e.g., sample size or method of randomization)

• Contradictory results just by chance

– Be concerned when discrepancies are large



• Do the results of the synthesis logically flow from the studies that were included? – Authors have a great deal of latitude in the

way they interpret and judge articles – Studies should be weighted based on study

size and validity• Reviews that simply compare the number of

positive findings to the number with negative findings may be misleading



• If the review dealt with therapy, was an estimate of treatment effect provided? – This is the overall estimate of treatment effect

size in a meta-analysis– The width of the associated confidence

interval reveals the precision of the estimate – A wide confidence interval degrades precision

and lowers the degree of certainty regarding the review’s findings



• Do the results of the review help with the care of patients? – In order to effectively use the results of a

review in the care of a specific patient, the populations should be similar

– The outcomes used in the review should be clinically important

– Comparable patient and review interventions• e.g., are manipulation and mobilization equal



• Were directions for future research offered? – Especially helpful in reviews because authors

assess the sum of all research on a topic and are in a very good position to determine what types of studies should be carried out in the future

© 2006 1 literature review designs narrative review systematic review meta-analysis

Documents

systematic literature

evidencebased chiropractic

literature review designs

evidence authors

types of literature

bias authors

highquality review

literature search authors