Evidence Synthesis and Critical Appraisal

Introduction

Health-care managers and evaluators routinely make decisions in conditions of uncertainty. They must determine whether an intervention improves outcomes, whether a service is safe and equitable, whether a diagnostic pathway is sufficiently accurate, and whether evidence generated elsewhere can guide local policy. Individual studies rarely answer these questions adequately. Studies may be too small, methodologically weak, inconsistent, or focused on outcomes that are not meaningful to patients or managers. Evidence synthesis provides a disciplined means of integrating findings, while critical appraisal assesses whether those findings are trustworthy and useful.

Evidence synthesis is more than collecting published articles. It involves formulating a decision-relevant question, searching systematically, selecting eligible evidence, assessing risk of bias, analysing or narratively integrating findings, judging certainty, and interpreting applicability. The process must also account for unpublished evidence, selective reporting, conflicts of interest, contextual differences, and the limits of research designs.

For postgraduate health-care managers, the central competency is not simply identifying the “best” study. It is making a transparent, proportionate and context-sensitive judgement about what is known, how confidently it is known, and what action is justified.

Learning objectives

By the end of this chapter, the reader should be able to:

1. Distinguish systematic, rapid, scoping, integrative and other forms of evidence review.
2. Translate a management or policy problem into a structured review question.
3. Design a reproducible search and screening process.
4. Appraise internal validity, applicability, precision and reporting integrity.
5. Explain meta-analysis, heterogeneity, publication bias and certainty of evidence.
6. Apply frameworks such as PRISMA, PICO, ROB 2, ROBINS-I and GRADE.
7. Adapt external evidence to local populations, systems, resources and values.
8. Recognize ethical, governance and conflict-of-interest issues in evidence synthesis.

1. Evidence synthesis as a decision process

A review is an argument supported by evidence, not merely a catalogue of references. Its credibility depends on alignment between the question, eligibility criteria, search strategy, appraisal method, synthesis method and conclusion.

The evidence hierarchy is useful but should not be treated as a rigid ladder. Randomized trials are generally strong for estimating comparative effects of interventions, but may be unsuitable for questions about long-term harms, organizational implementation, patient experience or rare events. Qualitative studies may provide the strongest evidence about acceptability and meaning. Administrative data may be valuable for utilization and equity analyses, although confounding and measurement error can be substantial. The appropriate design depends on the question.

A useful distinction is between:

• Internal validity: whether the study result is likely to be correct for the participants studied.
• Precision: how much statistical uncertainty surrounds the estimate.
• Applicability or external validity: whether the finding is relevant to the target population and setting.
• Certainty of evidence: confidence in the body of evidence for a particular outcome.
• Decision relevance: whether the evidence addresses outcomes, costs, feasibility and values that matter to the decision.

A highly precise estimate from a biased study is not trustworthy. A credible estimate from a setting unlike the target system may not be directly applicable. Evidence synthesis therefore requires both methodological and managerial judgement.

2. Major forms of evidence review

Systematic reviews

A systematic review uses pre-specified, explicit and reproducible methods to identify, select, appraise and synthesize all relevant evidence addressing a focused question. It may include meta-analysis, but a systematic review and a meta-analysis are not synonymous. Meta-analysis is a statistical method; a systematic review is a broader research process.

Systematic reviews are most appropriate when the question is sufficiently focused and the available evidence is mature enough to support structured comparison. They are commonly used to estimate intervention effects, diagnostic accuracy, prognosis or associations.

Rapid reviews

Rapid reviews streamline or omit selected systematic-review procedures to provide evidence within a shorter time. Examples include searching fewer databases, using one reviewer for an initial screen, limiting languages, or narrowing the question. These choices can be reasonable during an urgent policy decision, but they may increase the risk of missed studies, selection error and biased conclusions.

A rapid review should report its deviations from a full systematic review. Speed does not justify hiding methodological compromises. Decision-makers should know whether the review is fit for a high-stakes regulatory decision or only for preliminary orientation.

Scoping reviews

A scoping review maps the extent, range and nature of evidence on a broad topic. It can clarify concepts, identify evidence gaps, describe study designs and examine how a field has developed. Scoping reviews are useful when terminology is inconsistent or when the question is exploratory.

Unlike a systematic review focused on a specific effect estimate, a scoping review does not usually determine whether an intervention works. It may include diverse evidence and may not perform formal risk-of-bias assessment, although critical appraisal can be added where appropriate.

Other approaches

An umbrella review synthesizes existing systematic reviews. An integrative review may combine empirical, theoretical and methodological literature. A realist review asks what works, for whom, in what circumstances and why, making it valuable for complex interventions. A qualitative evidence synthesis integrates findings from qualitative studies to explain experiences, mechanisms and implementation conditions. A network meta-analysis compares multiple interventions, including comparisons not directly tested in the same trial, but relies on assumptions about transitivity and consistency.

3. Formulating the review question

A poorly formed question produces an inefficient search and an ambiguous conclusion. Frameworks help specify the decision problem.

For intervention questions, PICO identifies Population, Intervention, Comparator and Outcome. Variants include PICOT, which adds Time, and PICOS, which adds Study design. For qualitative questions, SPIDER—Sample, Phenomenon of Interest, Design, Evaluation and Research type—may be more suitable. For policy and implementation questions, frameworks may add setting, mechanism, feasibility, equity or resources.

A PICO question might be: Among adults with uncontrolled hypertension in primary care (Population), does pharmacist-led medication management (Intervention), compared with usual physician-led care (Comparator), improve blood-pressure control and reduce treatment burden (Outcomes) over 12 months (Time)?

Outcomes should be classified as critical, important or less important. Patient-important outcomes—mortality, quality of life, functional status, adverse events and treatment burden—should not be displaced by convenient surrogate outcomes. Managers may additionally require information about staffing, costs, waiting times, equity, implementation fidelity and sustainability.

Eligibility criteria should be established before screening. They should specify population, setting, intervention, comparator, outcomes, study designs, publication period, language and minimum follow-up where relevant. Criteria must be sufficiently precise to support consistent decisions, while avoiding unjustified exclusions that create selection bias.

4. Searching for evidence

A defensible search is sensitive enough to identify relevant evidence and specific enough to remain manageable. It commonly combines:

• Controlled vocabulary, such as database subject headings.
• Free-text synonyms, spelling variants and acronyms.
• Boolean operators: AND narrows, OR broadens, and NOT should be used cautiously.
• Citation chasing through reference lists and citing articles.
• Searches of trial registries, dissertations, government reports and professional organizations.
• Consultation with information specialists and subject experts.

The search strategy should be documented exactly, including databases, platforms, dates, search strings, filters and limits. A search conducted only in one biomedical database may miss economic, social-care, nursing, implementation or regional evidence.

Grey literature is important because commercially or institutionally unpublished evidence may differ from published evidence. However, grey sources vary in quality and should be assessed rather than automatically treated as unbiased. Searches should be updated near publication when the topic is rapidly changing.

A practical search workflow is:

1. Clarify the decision and structured question.
2. Identify key concepts and synonyms.
3. Test a preliminary search against known relevant studies.
4. Consult an information specialist.
5. Search multiple appropriate sources.
6. Record the complete strategy and date.
7. Search supplementary sources and update if necessary.

5. Screening, extraction and synthesis planning

Titles and abstracts are screened against the eligibility criteria, followed by full-text assessment. At least two independent reviewers are preferred for high-stakes reviews because independent checking reduces errors and subjective selection. Disagreements should be resolved through discussion or a third reviewer, with reasons recorded.

A PRISMA-style flow diagram reports the number of records identified, duplicates removed, records screened, full texts assessed, exclusions and included studies. The diagram improves transparency but does not itself guarantee methodological quality.

Data extraction forms should be piloted. Typical fields include study design, setting, sample, participant characteristics, intervention details, comparator, follow-up, outcomes, effect estimates, missing data, funding and author conflicts. Extracting outcome definitions and time points is essential because apparently similar outcomes may not be comparable.

A review protocol specifies the question, eligibility criteria, outcomes, search, appraisal and synthesis before results are known. Registration in a public repository, such as PROSPERO where eligible, reduces undisclosed changes and selective emphasis. Amendments may be justified, but they should be dated and explained.

6. Critical appraisal of individual studies

Critical appraisal asks whether a study is credible, interpretable and relevant. It should not be reduced to a numerical quality score. A study may be strong in one domain and weak in another; a total score can conceal important threats to validity.

Randomized trials

For randomized trials, appraisal includes:

• Adequacy of sequence generation and allocation concealment.
• Balance and comparability at baseline.
• Blinding of participants, clinicians, outcome assessors and analysts where feasible.
• Completeness of follow-up and handling of missing data.
• Adherence, contamination and co-interventions.
• Selective outcome reporting.
• Whether analysis follows the intention-to-treat principle.

The Cochrane ROB 2 tool organizes risk-of-bias judgements around randomization, deviations from intended interventions, missing outcome data, outcome measurement and selection of reported results.

Non-randomized studies

Observational studies are susceptible to confounding: differences between exposed and unexposed groups that influence both intervention receipt and outcome. Important issues include selection bias, time-varying exposure, immortal-time bias, inappropriate adjustment, misclassification and loss to follow-up. ROBINS-I evaluates risk of bias in non-randomized studies of interventions across domains including confounding, participant selection, intervention classification, deviations, missing data, outcome measurement and selective reporting.

Diagnostic studies

Diagnostic accuracy studies require appraisal of patient selection, index-test interpretation, reference-standard validity, timing and verification. Blinding is important because knowledge of the reference standard can influence interpretation of the index test. Spectrum bias occurs when study participants do not represent the range of patients encountered in practice.

Qualitative studies

Appraisal considers congruence between research question and methodology, sampling rationale, data collection, reflexivity, analytic transparency, researcher–participant relationships, ethical conduct and support for interpretations. A small sample is not automatically a weakness; adequacy depends on the research purpose and information richness.

Economic evaluations

Economic studies should be assessed for perspective, choice of alternatives, identification and measurement of costs and outcomes, valuation, time horizon, discounting, uncertainty and transferability. A cost-effective intervention from a societal perspective may not be affordable to a hospital operating under a narrow budget.

7. Quantitative synthesis and meta-analysis

When studies are sufficiently similar in participants, interventions, comparators and outcomes, their results may be statistically combined. The effect measure must match the outcome. Common measures include risk ratio, odds ratio, risk difference, hazard ratio, mean difference and standardized mean difference.

For a binary outcome, the risk ratio is:

Here, a is the number with the outcome in the intervention group, b is the number without it in that group, c is the number with the outcome in the comparator group, and d is the number without it in the comparator group. The formula assumes comparable definitions and follow-up, and it describes relative risk rather than absolute benefit.

The risk difference is:

It is expressed in percentage points or probability units. The number needed to treat, when the intervention is beneficial and the risk difference is non-zero, is:

The absolute value is used for magnitude; interpretation must still distinguish benefit from harm. NNT should be reported with its time horizon and outcome, because it changes with baseline risk and follow-up.

For continuous outcomes measured on the same scale, the mean difference is appropriate. When different instruments measure the same construct, the standardized mean difference may be used:

where M₁ and M₀ are intervention and comparator means and SDₚ is a pooled standard deviation. The assumptions include comparable constructs, meaningful scale direction and sufficiently similar populations. Standardized effects are less immediately interpretable to managers and patients.

A fixed-effect model assumes that all studies estimate one common underlying effect, with differences arising from sampling error. A random-effects model allows the true effect to vary across studies. Neither model solves clinical heterogeneity; statistical combination is inappropriate when interventions or outcomes are materially different.

Heterogeneity may be assessed using visual inspection, clinical and methodological comparison, and statistics such as I². The commonly used expression is:

where Q is Cochran’s heterogeneity statistic and df is its degrees of freedom. I² estimates the proportion of observed variation compatible with between-study heterogeneity rather than sampling error. It is influenced by study size and should not be interpreted using rigid cut-offs alone. A low I² does not prove that studies are clinically comparable, and a high I² does not identify its cause.

Subgroup and sensitivity analyses should be pre-specified where possible. Post hoc subgroup findings are exploratory. Meta-regression can examine whether study-level characteristics explain variation, but it is observational at the study level and vulnerable to ecological bias.

8. Publication bias and selective reporting

Publication bias occurs when the probability of publication depends on study results or their statistical significance. Related problems include outcome-reporting bias, time-lag bias, language bias, citation bias and sponsor-related dissemination bias. A synthesis of published studies may therefore overestimate benefits and underestimate harms.

Funnel plots can reveal asymmetry when enough studies are available, but asymmetry may also result from heterogeneity, small-study effects or methodological differences. Statistical tests for funnel-plot asymmetry have limited power with few studies and should not be treated as definitive. Searching registries, contacting authors, examining protocols, including grey literature and comparing registered outcomes with reported outcomes are important safeguards.

Selective reporting can occur at several levels: a study is not published, an outcome is omitted, a time point is changed, or an analysis is selected after examining the data. Reviewers should distinguish absence of evidence from evidence of absence. If harms are poorly reported, confidence in claims of safety should be reduced.

9. From risk of bias to certainty of evidence

Risk of bias concerns individual studies or specific results. Certainty concerns the body of evidence for a particular outcome and comparison. The GRADE approach commonly begins randomized evidence at high certainty and observational evidence at low certainty, then considers downgrading for risk of bias, inconsistency, indirectness, imprecision and publication bias. Evidence may be upgraded in selected circumstances, such as a large magnitude of effect, a dose–response gradient or residual confounding that would reduce rather than create the observed effect.

The four common certainty categories are:

• High: further research is unlikely to change confidence in the estimate.
• Moderate: further research may substantially change confidence.
• Low: confidence is limited and the true effect may differ substantially.
• Very low: the estimate is highly uncertain.

Certainty is outcome-specific. A review may provide moderate-certainty evidence for blood-pressure reduction, low-certainty evidence for quality of life and very-low-certainty evidence for long-term mortality. Conclusions should reflect this variation rather than assigning one overall label.

A Summary of Findings table presents the effect estimate, absolute risks, number of participants and studies, and reasons for downgrading. Absolute effects are particularly important for management decisions. For example, a large relative reduction may produce a small absolute benefit when baseline risk is low.

10. Applicability and adaptation to local contexts

External validity requires more than asking whether the local population resembles the research sample. Managers should examine:

1. Population: age, severity, comorbidity, socioeconomic position, language and health literacy.
2. Intervention: dose, components, workforce, training and fidelity.
3. Comparator: usual care may differ substantially between systems.
4. Setting: rural or urban location, facility level, referral pathways and infrastructure.
5. Baseline risk: absolute effects depend strongly on local risk.
6. Resources and costs: staff, equipment, medicines, data systems and opportunity costs.
7. Values and acceptability: patient preferences, cultural meanings and professional norms.
8. Equity: distribution of benefits, harms and access across disadvantaged groups.
9. Implementation capacity: leadership, workflow, regulation and sustainability.

A useful adaptation process is to separate transportability from implementation. Transportability asks whether the causal effect is likely to hold in the new context. Implementation asks whether the intervention can actually be delivered with adequate fidelity. An intervention may be effective but infeasible, or feasible but ineffective under local conditions.

Decision-makers should avoid simply importing a pooled estimate. They should examine baseline risk, intervention intensity, comparator quality and contextual mechanisms. Local pilot data, budget-impact analysis, stakeholder consultation and monitoring indicators can complement external evidence without pretending that a local pilot alone establishes causal effectiveness.

11. Conflicts of interest, ethics and governance

Conflicts of interest may be financial, professional, intellectual or institutional. They do not prove misconduct, but they can influence question selection, eligibility criteria, interpretation, publication and policy translation. Reviews should disclose funding, author relationships, sponsor involvement, protocol deviations and access to data.

A manufacturer-funded review may be methodologically rigorous, yet readers should examine whether unpublished trials were sought, whether harms were emphasized, and whether conclusions exceed the evidence. Review authors should use independent checking and, where possible, include contributors without relevant conflicts.

Evidence synthesis also raises ethical issues concerning privacy, secondary use of unpublished data, fair representation of marginalized populations and responsible communication of uncertainty. A review that excludes non-English evidence without justification may reproduce inequity. A review that communicates a weak association as a causal effect can cause harm through inappropriate policy.

Governance should include protocol review, data-management procedures, audit trails, transparent correction of errors and clear separation between evidence interpretation and organizational advocacy. Artificial-intelligence tools may assist deduplication, prioritization, translation, data extraction or screening, but human verification remains essential. AI systems can reproduce training-data bias, misclassify eligibility, fabricate citations or obscure accountability. Their use should be documented, validated and proportionate to risk.

12. Applied case: evaluating virtual hypertension management

Background

The fictional Riverland Health Network is considering a virtual hypertension-management service involving home blood-pressure monitoring, pharmacist review and telephone or video consultations. Senior managers want to know whether the service should be implemented across 24 primary-care clinics.

Problem

A preliminary search identifies several randomized trials, observational evaluations and qualitative studies. Trial results appear favorable, but interventions differ in monitoring frequency, pharmacist authority, digital access and usual care. The network serves older adults, rural communities and patients with limited internet access.

Analysis

The evaluation team formulates a PICO question: in adults with uncontrolled hypertension receiving primary care, does pharmacist-supported remote monitoring compared with usual care improve blood-pressure control, adverse events, patient experience and equity over at least six months?

The team conducts a systematic search of biomedical, nursing, health-services and economic databases, trial registries and government sources. Two reviewers independently screen studies. They include randomized and comparative observational studies, but synthesize them separately. The protocol specifies systolic blood pressure, serious adverse events, treatment burden, medication adherence and equity as outcomes.

Risk of bias is assessed using ROB 2 for randomized trials and ROBINS-I for non-randomized studies. The trials show a modest average reduction in systolic blood pressure, but certainty is downgraded for inconsistency and indirectness because most studies use younger, digitally confident populations. Evidence on mortality, equity and long-term sustainability is sparse. Qualitative studies identify improved reassurance for some patients but anxiety and device burden for others.

The team examines baseline risk and local capacity. Riverland has pharmacists but limited broadband in rural areas. A telephone option is therefore more feasible than a smartphone-only platform. A budget-impact analysis includes devices, staff time, training, technical support and replacement costs. Equity analysis identifies the possibility that remote monitoring could widen disparities unless devices are loaned, interpretation is available and patients can choose telephone or in-person care.

Approach and solution

Managers approve a staged implementation rather than immediate universal adoption. Twelve clinics begin a 12-month pragmatic evaluation, with matched comparison clinics where randomization is not feasible. Pre-specified outcomes include blood-pressure control, medication-related harm, missed appointments, patient-reported burden, uptake by socioeconomic group, staff workload and cost per controlled patient. A stopping rule is established for serious safety signals.

Outcome

The decision is not based solely on the pooled effect. It combines moderate-certainty evidence for a clinical outcome with low certainty for equity and sustainability, local feasibility information and patient preferences. The programme is adapted to local infrastructure and monitored for unintended consequences.

Learning points

1. A pooled estimate does not automatically transfer across settings.
2. Outcomes important to managers include feasibility, workload, cost and equity.
3. Mixed evidence designs can be synthesized, but unlike designs should not be casually combined.
4. Staged implementation can manage uncertainty while generating local learning.
5. Adaptation should preserve the intervention’s plausible mechanism while changing its delivery format where necessary.

13. Best-practice checklist

Before relying on a review, ask:

• Is the question explicit and decision relevant?
• Was the protocol registered or otherwise documented?
• Was the search broad, reproducible and recently updated?
• Were study selection and extraction independently checked?
• Were unpublished evidence and conflicts considered?
• Was risk of bias assessed with a design-appropriate tool?
• Are synthesis methods justified by clinical and methodological similarity?
• Are absolute effects and uncertainty reported?
• Is certainty judged separately for important outcomes?
• Are applicability, equity, feasibility and local baseline risk addressed?
• Do conclusions avoid stronger claims than the evidence supports?

Chapter Summary

Evidence synthesis is a structured process for answering health-care questions by identifying, appraising and integrating relevant research. Systematic reviews address focused questions through reproducible methods; rapid reviews trade some methodological completeness for timeliness; and scoping reviews map concepts and evidence without necessarily estimating effectiveness. A strong review begins with a precise question, comprehensive search, transparent eligibility criteria, independent screening, careful extraction and a prespecified protocol.

Critical appraisal examines internal validity, precision, applicability and decision relevance. Design-specific tools such as ROB 2 and ROBINS-I help identify bias, while GRADE assesses certainty across the body of evidence. Meta-analysis can improve precision when studies are sufficiently comparable, but statistical pooling cannot correct bias or eliminate clinical heterogeneity. Publication bias, selective reporting and conflicts of interest may distort the evidence base.

External evidence must be adapted to local populations, baseline risks, resources, infrastructure, values, equity concerns and implementation capacity. Effective health-care evaluation therefore combines research synthesis with contextual analysis, stakeholder engagement, economic reasoning and monitored implementation. The most defensible decision is not necessarily the one supported by the largest number of studies, but the one that transparently aligns trustworthy evidence with the realities and values of the population served.

Key Terms

Applicability: Relevance of evidence to a particular population, setting or decision.

Bias: Systematic deviation from the truth caused by design, conduct, analysis or reporting.

Certainty of evidence: Confidence that an estimated effect or conclusion is close to the truth.

Clinical heterogeneity: Differences among studies in participants, interventions, comparators or outcomes.

Conflict of interest: A relationship or circumstance that could improperly influence judgement or interpretation.

Critical appraisal: Systematic assessment of a study’s credibility, relevance and limitations.

External validity: Extent to which findings can be generalized or transported to other settings.

GRADE: Framework for judging certainty of evidence and strength of recommendations.

Grey literature: Relevant material not distributed through conventional commercial publication channels.

I²: Statistic describing the proportion of observed variability compatible with between-study heterogeneity.

Meta-analysis: Statistical combination of results from multiple studies.

PICO: Population, Intervention, Comparator and Outcome framework for focused questions.

Publication bias: Distortion caused by selective publication of studies or results.

Rapid review: Evidence review using abbreviated methods to provide timely findings.

Risk of bias: Likelihood that study methods systematically distort an estimate.

Scoping review: Review that maps the extent, characteristics and gaps in a body of evidence.

Systematic review: Review using explicit, reproducible methods to identify and synthesize relevant evidence.

Transportability: Likelihood that a causal effect applies in a different context.

Review Questions

Short Answer Questions

1. What is the difference between a systematic review and a meta-analysis?
2. Why should patient-important outcomes be specified in a review protocol?
3. What does risk of bias assess, and how does it differ from certainty of evidence?
4. Identify two potential sources of publication bias.
5. Why may a statistically significant relative effect have limited managerial importance?

Long Answer Questions

1. Explain the stages of a systematic review from question formulation through interpretation, and discuss how errors at each stage may affect conclusions.
2. Compare systematic, rapid and scoping reviews in terms of purpose, methods, strengths and limitations.
3. Discuss how ROB 2, ROBINS-I and GRADE contribute different forms of judgement in evidence evaluation.
4. Explain why meta-analysis may be inappropriate even when several studies address the same broad intervention.
5. Describe a framework for adapting external evidence to a local health-care system while protecting equity and patient safety.

Analytical / Application Questions

1. A review reports a large pooled reduction in hospital admissions, but includes mostly retrospective studies funded by the intervention manufacturer and provides little information about missing data. How would you appraise the credibility of the conclusion?
2. A digital-health intervention is effective in urban tertiary hospitals but is being considered in rural clinics with limited broadband and few specialists. What evidence and contextual factors should managers examine before adoption?
3. Design a review protocol for evaluating a nurse-led chronic-disease management programme. Specify the question framework, eligibility criteria, outcomes, search sources, risk-of-bias approach and method for judging certainty.

Multiple Choice Questions with Answers

Question 1

Which feature most clearly distinguishes a systematic review from a narrative literature review?

A. It includes only randomized controlled trials.

B. It uses explicit, reproducible methods for searching, selecting and appraising evidence.

C. It always performs a meta-analysis.

D. It excludes qualitative research.

Correct Answer: B. It uses explicit, reproducible methods for searching, selecting and appraising evidence.

Question 2

What is the primary purpose of a scoping review?

A. To calculate one definitive treatment effect.

B. To replace regulatory approval studies.

C. To map the extent, range and characteristics of evidence on a topic.

D. To prove that an intervention is cost-effective.

Correct Answer: C. To map the extent, range and characteristics of evidence on a topic.

Question 3

In the PICO framework, the letter C represents:

A. Certainty

B. Context

C. Comparator

D. Confounding

Correct Answer: C. Comparator

Question 4

Which issue is specifically assessed when a review examines whether study results were selectively published?

A. Publication bias

B. Allocation concealment

C. Spectrum bias

D. Recall bias

Correct Answer: A. Publication bias

Question 5

If intervention risk is 10% and comparator risk is 20%, the risk difference is:

A. −30 percentage points

B. −10 percentage points

C. 10 percentage points

D. 50 percentage points

Correct Answer: B. −10 percentage points

Question 6

Which statement about I² is most accurate?

A. It proves that a pooled result is unbiased.

B. It measures the proportion of observed variability compatible with between-study heterogeneity.

C. It identifies which specific study caused heterogeneity.

D. It can replace clinical judgement about study comparability.

Correct Answer: B. It measures the proportion of observed variability compatible with between-study heterogeneity.

Question 7

Under GRADE, which is a common reason for downgrading certainty?

A. A clearly stated research question

B. Use of patient-important outcomes

C. Serious risk of bias or imprecision

D. Registration of the review protocol

Correct Answer: C. Serious risk of bias or imprecision

Question 8

Which tool is designed primarily for assessing risk of bias in randomized trials?

A. ROB 2

B. ROBINS-I

C. SPIDER

D. PRISMA

Correct Answer: A. ROB 2

Question 9

Why should external evidence be examined against local baseline risk?

A. Relative effects never transfer between settings.

B. Absolute benefits and harms depend partly on baseline risk.

C. Baseline risk determines whether randomization occurred.

D. It eliminates the need for local stakeholder consultation.

Correct Answer: B. Absolute benefits and harms depend partly on baseline risk.

Question 10

Which is the most appropriate use of artificial intelligence in evidence synthesis?

A. Allowing an unverified system to make final eligibility decisions in a high-stakes review.

B. Using AI for prioritization or deduplication with documented human verification.

C. Replacing conflict-of-interest disclosure with automated classification.

D. Accepting AI-generated citations without checking their existence.

Correct Answer: B. Using AI for prioritization or deduplication with documented human verification.