Академический Документы
Профессиональный Документы
Культура Документы
SUMMARISING AND
Prathap Tharyan, MD, MRCPsych
INTERPRETING FINDINGS USING Adjunct Professor, Clinical
Epidemiology Unit
THE GRADE APPROACH
Grading of Recommendations: Assessment,
Development and Evaluation
http://www.gradeworkinggroup.org
ADDRESSING THE GAPS BETWEEN EVIDENCE AND PRACTICE
BVMC |
CMCVellore
BVMC | 4
CMCVellore
FOREST PLOT OF COMPARISON: PRIMAQUINE 5 DAYS VERSUS PRIMAQUINE 14 DAYS
OUTCOME: 1.1P. VIVAXPARASITAEMIA > 30 DAYS AFTER STARTING PRIMAQUINE.
BVMC | 5
CMCVellore
RISK OF BIAS
BVMC | 6
CMCVellore
BVMC | 7
CMCVellore
FOREST PLOT OF COMPARISON: 1 PRIMAQUINE: 5 DAYS VERSUS 14 DAYS,
OUTCOME: 1.1P. VIVAXPARASITAEMIA > 30 DAYS AFTER STARTING
PRIMAQUINE.
BVMC | 8
CMCVellore
SUMMARIZING THE RESULTS OF A SYSTEMATIC REVIEW:
• In order to evaluate the results of a systematic review to inform policy
decisions, we need to know:
1. Do the pooled effect estimates (with 95% CIs) for each clinically
important outcome indicate appreciable advantages?
• Efficacy | Safety | Adherence | Costs | Quality of life | Other
2. How confident are we that these effects are likely to be seen in
practice? How certain are we that further research will not
change these estimates?
• A separate summary of findings table is created for each comparison
(if more than one)
• For each outcome (maximum 7 clinically important outcomes) the
GRADE quality (confidence / certainty) assessment is presented along
with effect estimate and absolute difference (with 95% CIs)
BVMC |
CMCVellore
CONFIDENCE IN THE RESULTS DEPENDS ON THE QUALITY OF
EVIDENCE
Can we trust the evidence?
The risk of bias of the included trials: (Study Limitations)
Does the evidence apply to my setting?
The characteristics of the individual trials:
Population/comparisons/outcomes/health systems (Directness)
Are the results consistent in all the trials?
If not, can heterogeneity be explained? (Inconsistency)
Are the results precise?
Do the results clearly indicate statistically significant and clinically important
effects ? (Imprecision)
Would your treatment decision change if the effect estimate was at either
end of the 95% CI?
Are the data sufficient based on populations covered and number of events?
Is there evidence complete?
Did the review include all relevant trials or only easily accessed ones?
(Publication bias)
BVMC |
CMCVellore
BVMC |
CMCVellore
BVMC |
CMCVellore
Linking numerical results to our confidence in the effect
estimates
Levels of certainty or uncertainty Terminology
4 = High We are very confident that the true effect lies close to that
RCTs of the estimate of the effect; It will
Further research is unlikely to change this estimate
1 = Very low We have very little confidence in the effect estimate: The
true effect is likely to be substantially different from the
estimate of effect; further research is needed and is very
We don’t
likely to alter these estimates know
BVMC |
CMCVellore
USING THE EVIDENCE: TO INFORM POLICY DECISIONS
BVMC |
CMCVellore
RATING QUALITY OF
EVIDENCE
Study design
STUDY DESIGN
• In the GRADE approach to quality of evidence:
• randomized trials without important limitations provide
high quality evidence
• observational studies without special strengths or
important limitations provide low quality evidence
• Limitations or special strengths can, however, modify
the quality of the evidence of both randomized trials
and observational studies.
BVMC |
CMCVellore
STUDY DESIGN
BVMC |
CMCVellore
RATING QUALITY OF EVIDENCE
RCTs
For RCTs:
5 criteria to downgrade
• Limitations (of study design)
• Inconsistency (between results)
• Indirectness (of evidence)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
LIMITATIONS IN STUDY QUALITY
• RCTs
• Improper generation of randomization sequence
• Lack of allocation concealment
• Lack of blinding leading to performance or detection bias
• Incomplete accounting for patients and outcomes
• Selective reporting or publication bias
• Other limitations
• stopping early for benefit observed in randomized trials, in particular in the
absence of adequate stopping rules
• use of un-validated patient-reported outcomes
• carry-over effects in cross-over trials
• recruitment bias in cluster-randomized trials
BVMC |
CMCVellore
The COCHRANE RISK OF BIAS TOOL
• Risk of bias table
• Risk of bias graph
• Risk of bias summary figure
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) No serious inconsistency
Serious inconsistency (-1)
Very serious inconsistency (-2)
• Indirectness (of evidence)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
STATISTICAL HETEROGENEITY
• Statistical heterogeneity exists when the results of individual
studies are not “consistent” among themselves.
Clinical diversity
Methodological diversity Statistical
Biases heterogeneity
Chance
BVMC |
CMCVellore
IDENTIFYING IMPORTANT INCONSISTENCY
• Non overlapping confidence intervals
• Chi-squared test with P value < 0.1
• I2 : I2 of 25% = low; 50% = moderate and 75% = high (tentative)
Higgins et al, BMJ 2003; 327: 557-560
Current interpretation:
• 0% to 40%: might not be important;
• 30% to 60%: may represent moderate heterogeneity*;
• 50% to 90%: may represent substantial heterogeneity*;
• 75% to 100%: considerable heterogeneity*.
• *The importance of the observed value of I2 depends on :
• magnitude and direction of effects and
• strength of evidence for heterogeneity (e.g. P value from the chi-
squared test, or a confidence interval for I2).
BVMC |
CMCVellore
IMPORTANT INCONSISTENCY
• Unexplained heterogeneity may be a reason to downgrade
• If heterogeneity is explained (like in sub-group analyses),
then one need not downgrade
• Unexplained heterogeneity may or may not be a reason to
downgrade when results are variable but the differences are
between small and large beneficial effects
BVMC |
CMCVellore
Is there important heterogeneity ?
BVMC |
CMCVellore
SHOULD WE DOWNGRADE?
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) Serious inconsistency (-1)
• Indirectness (of evidence)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
INDIRECTNESS
• Population: All studies includes only adults when children are also
affected in practice
• Intervention: Given in doses or ways not used in practice
• Comparator: Given in doses or ways not used in practice (E.g.: Newer
antipsychotics compared against high dose haloperidol and not
variable/low dose)
• Outcome: Not defined in similar ways, not assessed in all trials;
surrogate outcomes used. (E.g.: Rosiglitazone produces tight glycaemic
control but no evidence on preventing complications)
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) Serious inconsistency (-1)
• Indirectness (of evidence) Very serious indirectness (-2)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
GRADING IMPRECISION: I- (95% CIS)
No serious imprecision Both limits of the 95% CI of the pooled estimate suggest
appreciable benefit* with one drug
Serious The 95% CI of the pooled estimate includes appreciable
benefit* and non-appreciable benefit$ with one drug
BVMC |
CMCVellore
SERIOUS IMPRECISION
BVMC |
CMCVellore
VERY SERIOUS IMPRECISION
BVMC |
CMCVellore
GRADING IMPRECISION I: (95% CIS)
Exception:
• When event rates are very low:
• 95% confidence intervals around relative effects
(RR) can be very wide, but
• 95% confidence intervals around absolute effects
(RD) may be narrow.
• Under such circumstances:
• One need not downgrade the quality of evidence for
imprecision if the sample size is large (2000 in each
arm)
BVMC |
CMCVellore
GRADING IMPRECISION: II
SAMPLE SIZE AND OPTIMAL INFORMATION SIZE
(OIS)
• Other reasons to downgrade for imprecision:
• Rule of thumb: Total number of events is less than 300
OR
• Total (cumulative) sample size is lower than the calculated optimal
information size (OIS)
BVMC |
CMCVellore
OPTIMAL INFORMATION SIZE
• Optimal information size (OIS): represents the
number of patients generated by a conventional
sample size calculation specifying a particular
alpha and beta error, relative risk reduction, and
baseline event rate
BVMC |
CMCVellore
BVMC |
CMCVellore
OPTIMAL INFORMATION SIZE GIVEN Α OF 0.05 AND Β OF 0.2 FOR VARYING CONTROL EVENT
RATES AND RELATIVE RISKS
BVMC | 42
Journal of Clinical Epidemiology 2011 64, 1283-1293DOI: (10.1016/j.jclinepi.2011.01.012) CMCVellore
IMPRECISION:
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) Serious inconsistency (-1)
• Indirectness (of evidence) Very serious indirectness (-2)
• Imprecision (of measure of effect) (No serious imprecision)
• Other (publication bias)
BVMC |
CMCVellore
Reporting biases
BVMC |
CMCVellore
We would expect less precise studies (with fewer participants and events) to be
more affected by the play of chance, and so more widely scattered about the
pooled estimate.
As studies get bigger with more events, we expect them to be closer to the
pooled estimate.
Overall, this should produce a triangular shape, or inverted funnel (depending on
how the axes are plotted)
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) Serious inconsistency (-1)
• Indirectness (of evidence) Very serious indirectness (-2)
• Imprecision (of measure of effect) (No serious imprecision)
• Other (publication bias) (No other biases)
BVMC |
CMCVellore
RATING QUALITY OF EVIDENCE
Observational studies
• Controlled before and after studies
• Interrupted time series
• Cohort studies
4 levels of ‘Quality’
• 4 = High
• 3 = Moderate
Observational
• 2 = Low
• 1 = Very low
BVMC |
CMCVellore
OBSERVATIONAL STUDIES
• Observational studies start as low quality
• Those that have been downgraded for any reason, as well as,
• Uncontrolled case series starting as very low quality
• Cannot be upgraded.
BVMC |
CMCVellore
LIMITATIONS TO QUALITY IN OBSERVATIONAL
STUDIES: 1
• Failure to develop and apply appropriate eligibility criteria
(inclusion of control population)
• under- or over-matching in case-control studies
• selection of exposed and unexposed in cohort studies from different
populations
Observational
BVMC |
CMCVellore
LIMITATIONS TO QUALITY IN OBSERVATIONAL
STUDIES: 2
• Flawed measurement of both exposure and outcome
• differences in measurement of exposure
• recall bias in case- control studies
• differential surveillance for outcome in exposed and unexposed in
cohort studies
BVMC |
CMCVellore
LIMITATIONS TO QUALITY IN OBSERVATIONAL
STUDIES: 3
• Failure to adequately control confounding
• failure of accurate measurement of all known prognostic
factors
• failure to match for prognostic factors and/or adjustment in
statistical analysis
• Incomplete or inadequately short follow-up
• Downgrade 1 or 2 levels depending on extent to which
limitations may bias results
BVMC |
CMCVellore
3 additional criteria: to upgrade evidence from
Observational studies
• Strong association
• Confounders all act to reduce observed effect
• Dose-response effect
BVMC |
CMCVellore
UPGRADING OBSERVATIONAL
STUDIES
• For magnitude of effect:
• Large effect:
• A meta-analysis of observational studies (with no threats to validity)
showed that bicycle helmets reduce the risk of head injuries in cyclists
(OR: 0.31, 95%CI: 0.26 to 0.37).
• This large effect suggests the evidence could be upgraded one level
• Very large effect:
• A meta-analysis of observational studies (with no limitations) showed that
warfarin prophylaxis reduces the risk of thromboembolism in patients with
cardiac valve replacement (RR: 0.17, 95%CI: 0.13 to 0.24).
• This very large effect suggests the evidence could be upgraded two levels
BVMC |
CMCVellore
UPGRADING OBSERVATIONAL
STUDIES
• For plausible confounding:
• if only sicker patients receive an experimental
intervention or exposure, yet they still fare better,
it is likely that the actual intervention or exposure
effect is larger than the data suggest.
• Upgrade 1 level
BVMC |
CMCVellore
UPGRADING OBSERVATIONAL
STUDIES
• A dose-response gradient:
• In observational studies (with no study
limitations) in patients receiving anticoagulation
with warfarin, higher levels of the international
normalized ratio (INR) leads to increased risk of
bleeding
• This increases our confidence that supra-
therapeutic anticoagulation levels increase
bleeding risk.
• Upgrade 1 level
BVMC |
CMCVellore
BVMC |
CMCVellore
SOME VARIATIONS
BVMC |
CMCVellore
DIFFERENCES FOR CONTINUOUS OUTCOMES
BVMC |
CMCVellore
DIFFERENCES FOR CONTINUOUS OUTCOMES
BVMC |
CMCVellore
DIFFERENCES FOR CONTINUOUS OUTCOMES
BVMC |
CMCVellore
Part ii:
Primaquine to
prevent relapse
with vivax
malaria
Summary of findings
BVMC |
CMCVellore
FIVE DAYS VERSUS 14 DAYS OF PRIMAQUINE + CHLOROQUINE
BVMC |
CMCVellore
FIVE DAYS VERSUS 14 DAYS OF PRIMAQUINE +
CHLOROQUINE
BVMC |
CMCVellore
STRENGTH OF RECOMMENDATIONS
BVMC |
CMCVellore
Part 3: Exercise-
Grading
evidence for
guidelines
Falciparum Malaria
BVMC |
CMCVellore
BVMC |
CMCVellore
BVMC |
CMCVellore
BVMC |
CMCVellore
BVMC |
CMCVellore
PICO QUESTION
In patients diagnosed with severe falciparum malaria does treatment
with Artesunate (i.v.) reduce death compared to treatment with Quinine
(i.v.)?
BVMC |
CMCVellore
BVMC |
CMCVellore
ARTESUNATE VERSUS QUININE IN SEVERE MALARIA
OUTCOME: DEATH
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death
Effect estimate:
Relative = RR 0.62 (95% CI 0.51 to 0.75)
Absolute = ARR = 84 fewer per 1,000 (95% CI 56 to 109)
NNT = 1/ARR = 12 (95% 9 to 18)
BVMC |
CMCVellore
PART III: CAN WE TRUST THESE
EFFECT ESTIMATES?
BVMC |
CMCVellore
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results)
• Indirectness (of evidence)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
INCONSISTENCY
BVMC |
CMCVellore
AL6 versus AQ+SP in uncomplicated malaria
Outcome: Treatment failure at day 28
BVMC |
CMCVellore
Pneumococcal vaccine versus Placebo
Outcome: X-ray defined pneumonia
Effect estimate:
RR 0.79 (95% CI 0.73 to 0.85)
BVMC |
CMCVellore
Pneumococcal vaccine versus Placebo
Outcome: X-ray defined pneumonia
Effect estimate:
RR 0.82 (95% CI 0.68 to 1.00)
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) No serious inconsistency
• Indirectness (of evidence)
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
INDIRECTNESS
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death (subgrouped by age)
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death (subgrouped by IV/IM artesunate)
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death (sub-grouped by loading dose Quinine)
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) No serious inconsistency
• Indirectness (of evidence) No serious indirectness
• Imprecision (of measure of effect)
• Other (publication bias)
BVMC |
CMCVellore
IMPRECISION
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death
A precise result?
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Neurological disability at discharge
A precise result?
BVMC |
CMCVellore
Pneumococcal vaccine versus Placebo
Outcome: Clinical pneumonia
A precise result?
BVMC |
CMCVellore
Pneumococcal vaccine versus Placebo
Outcome: X-ray defined pneumonia
A precise result?
BVMC |
CMCVellore
DHA-P versus AS+MQ in uncomplicated malaria
Outcome: Serious adverse events
An imprecise result?
BVMC |
CMCVellore
5 criteria
• Limitations (of study design) No serious limitations
• Inconsistency (between results) No serious inconsistency
• Indirectness (of evidence) No serious indirectness
• Imprecision (of measure of effect) No serious imprecision
• Other (publication bias)
BVMC |
CMCVellore
OTHER
BVMC |
CMCVellore
Standard Error 1
0.5 1.0 2
Risk ratio
BVMC |
CMCVellore
Artesunate versus quinine in severe malaria
Outcome: Death
BVMC |
CMCVellore
EVIDENCE PROFILES
BVMC |
CMCVellore
Artesunate compared to Quinine for Adults with severe malaria
Patient or population: Adults with severe malaria
Settings: India
Intervention: Artesunate
Comparison: Quinine
Outcomes Illustrative comparative risks* (95% CI) Relative effect Participants Quality Comments
(95% CI) (studies) (GRADE)
Assumed risk Corresponding risk
Quinine Artesunate
Death 222 per 1000 84 fewer per 1000 RR 0.62 1938 ⊕⊕⊕⊕
(from 56 fewer to (0.51 to 0.75) (6 studies) high1,2
109 fewer)
Neurological 5 per 1000 6 more per 1000 RR 2.21 1190 ⊕⊕⊝⊝
sequelae at (from 2 fewer to 35 (0.64 to 7.63) (2 studies) low3,4,5
discharge more)
Coma recovery The mean The mean Coma 231 ⊕⊝⊝⊝
time coma recovery recovery time in the (2 studies) very low
time in the intervention groups 6,7,8
Death 222 per 1000 84 fewer per 1000 RR 0.62 1938 ⊕⊕⊝⊝
(from 56 fewer to (0.51 to 0.75) (6 studies) low1,2
109 fewer)
Neurological 5 per 1000 6 more per 1000 RR 2.21 1190 ⊕⊝⊝⊝
Sequelae (from 2 fewer to 35 (0.64 to 7.63) (2 studies) very low3,4,5
more)
Coma recovery The mean The mean Coma 231 ⊕⊝⊝⊝
time coma recovery recovery time in the (2 studies) very low6,7,8
time in the intervention groups
control groups Was 2.11 hours
Was 55.80 Longer (from 19.17
Hours hrs quicker to 23.4
hrs longer)
Time to hospital The mean time The mean Time to 113 ⊕⊝⊝⊝
discharge to hospital hospital discharge (1 study) very low5, 9,10,
discharge in in the intervention
the control groups was
groups was 0.10 days longer
5.00 Days (1.34 days quicker BVMC |
to 1.54 days CMCVellore
Would you recommend Artemisinin in adults?
Would you recommend Artemisinin in children?
BVMC |
CMCVellore
BVMC |
CMCVellore
THE ROLE OF EVIDENCE
Local evidence
Systematic reviews
Judgements about modifying
Judgements about
factors, needs, values,
the impact of
resources
policies
BVMC |
CMCVellore
“When you know something, to hold
that you know it, and when you do
not know something, to allow that
you do not know it; that is
knowledge.”
Confucius (551 BC - 479 BC)
BVMC |
CMCVellore