2. What is EBM?
ââThe conscientious, explicit and judiciousThe conscientious, explicit and judicious
use ofuse of current best evidencecurrent best evidence inin
making decisions about the care ofmaking decisions about the care of
individual patientsâindividual patientsâ
Prof. David L. Sackett, 1997Prof. David L. Sackett, 1997
Dr Max Mongelli 2016
5. Primum non nocerePrimum non nocere
ââFirst do no harmâFirst do no harmâ
Hippocrates,Hippocrates, EpidemicsEpidemics
Dr Max Mongelli 2016
6. Sources of Evidence in Medicine
⢠Traditional TeachingTraditional Teaching
⢠TextbooksTextbooks
⢠Basic sciencesBasic sciences
⢠Observational studiesObservational studies
⢠Computer simulationComputer simulation
⢠Decision AnalysisDecision Analysis
⢠Case-Control StudiesCase-Control Studies
⢠Randomised Controlled Trials (RCT)Randomised Controlled Trials (RCT)
⢠Meta-analysesMeta-analyses
Dr Max Mongelli 2016
7. RCOG Classification of Evidence LevelsRCOG Classification of Evidence Levels
ďŽ 1++ High quality meta-analyses of RCT's1++ High quality meta-analyses of RCT's
ďŽ 1+ Meta-a. Or RCT's at low risk of bias1+ Meta-a. Or RCT's at low risk of bias
ďŽ 1- Meta-a. Or RCT's at high risk of bias1- Meta-a. Or RCT's at high risk of bias
ďŽ 2++ High quality meta-analyses of CC's2++ High quality meta-analyses of CC's
ďŽ 2+ Well-conducted cc or cohort studies2+ Well-conducted cc or cohort studies
ďŽ 2- Case-control or cohort studies with ?2- Case-control or cohort studies with ?
biasbias
ďŽ 3 Case reports3 Case reports
ďŽ 4 Expert opinion4 Expert opinion Dr Max Mongelli 2016
8. ďŽ ââEffectiveness andEffectiveness and
Efficiency: RandomEfficiency: Random
Reflections of HealthReflections of Health
ServicesServices â, 1971â, 1971
Archie Cochrane (1909-88)Archie Cochrane (1909-88)
Dr Max Mongelli 2016
9. Randomized Controlled TrialsRandomized Controlled Trials
ďŽ ââGold standardâ in evaluating newGold standardâ in evaluating new
therapies or surgical techniquestherapies or surgical techniques
ďŽ May also be applied to new diagnosticMay also be applied to new diagnostic
teststests
Dr Max Mongelli 2016
10. Objectives of RCT:Objectives of RCT:
ďŽ Minimize bias by randomisationMinimize bias by randomisation
ďŽ Achieve statistical power throughAchieve statistical power through
adequate sample sizeadequate sample size
ďŽ ââBlinding â - single or doubleBlinding â - single or double
ďŽ Analysis by intention to treatAnalysis by intention to treat
Randomized Controlled TrialsRandomized Controlled Trials
Dr Max Mongelli 2016
11. Randomized Controlled TrialsRandomized Controlled Trials
RandomisationRandomisation
ďŽ Several techniques availableSeveral techniques available
ďŽ Computer software linked to centralComputer software linked to central
monitoring stationmonitoring station
ďŽ ââBlock â randomisationBlock â randomisation
ďŽ Sealed envelope methodSealed envelope method
Dr Max Mongelli 2016
14. RCTâs and Observational StudiesRCTâs and Observational Studies
⢠Two studies published in the NEJM in 2000 suggested that
RCTs and observational studies overall produced similar
results
⢠JAMA 2001: âdiscrepancies beyond chance do occur and
differences in estimated magnitude of treatment effect are
very commonâ
⢠RCTs may be unnecessary for treatments that have
dramatic and rapid effects relative to the expected
Dr Max Mongelli 2016
15. RCTâs and Industry FundingRCTâs and Industry Funding
â˘RCTâs funded by industry are significantly more likely to
report positive results
â˘Possibly due to publication bias
â˘RCTs may be unnecessary for treatments that have
dramatic and rapid effects relative to the expected
Dr Max Mongelli 2016
16. RCTâs and Statistical ErrorRCTâs and Statistical Error
⢠Type I error â âfalse positiveâ
⢠Type II error â âfalse negativeâ
⢠Sample size calculations often inaccurate
Dr Max Mongelli 2016
18. 2 X 2 Table2 X 2 Table
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
19. 2 X 2 Table2 X 2 Table
ďŽ Sensitivity = TP =Sensitivity = TP =
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
20. 2 X 2 Table2 X 2 Table
ďŽ Sensitivity = TP = a/(a+c)Sensitivity = TP = a/(a+c)
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
21. 2 X 2 Table2 X 2 Table
ďŽ False Positive Rate = FP =False Positive Rate = FP =
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
22. 2 X 2 Table2 X 2 Table
ďŽ FP = b/(d+b)FP = b/(d+b)
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
23. 2 X 2 Table2 X 2 Table
ďŽ Specificity = 1 - FP =Specificity = 1 - FP =
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
24. 2 X 2 Table2 X 2 Table
ďŽ Specificity = 1 - FP = d/(d+b)Specificity = 1 - FP = d/(d+b)
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
25. 2 X 2 Table2 X 2 Table
ďŽ Positive predictive value (PPV) =Positive predictive value (PPV) =
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
26. 2 X 2 Table2 X 2 Table
ďŽ PPV = a/(a+b)PPV = a/(a+b)
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
27. 2 X 2 Table2 X 2 Table
ďŽ Negative predictive value (NPV)Negative predictive value (NPV)
==
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
28. 2 X 2 Table2 X 2 Table
ďŽ NPV = d/(c+d)NPV = d/(c+d)
Disease
present
Disease
absent
Test
Positive
a b
Test
Negative
c d
Dr Max Mongelli 2016
29. PPV and PrevalencePPV and Prevalence
ďŽ Steep drop in positive predictiveSteep drop in positive predictive
value as disease prevalencevalue as disease prevalence
decreasesdecreases
Dr Max Mongelli 2016
30. PPV and PrevalencePPV and Prevalence
ďŽ PPV =PPV = (sens x prev)(sens x prev)
(sens x prev +(1 - spec)x(1 - prev))(sens x prev +(1 - spec)x(1 - prev))
Dr Max Mongelli 2016
31. The Likelihood RatioThe Likelihood Ratio
ďŽ Single value to indicate theSingle value to indicate the
clinical utility of a testclinical utility of a test
ďŽ Independent of prevalenceIndependent of prevalence
ďŽ LR = Sensitivity/(1- Spec.)LR = Sensitivity/(1- Spec.)
ďŽ LR >8 : tests usuallyLR >8 : tests usually
clinically usefulclinically useful
Dr Max Mongelli 2016
32. The Likelihood RatioThe Likelihood Ratio
ďŽ LR is an odds modifier:LR is an odds modifier:
ďŽ Posterior odds =Posterior odds =
prior odds x LRprior odds x LR
Dr Max Mongelli 2016
33. Odds and ProbabilityOdds and Probability
ďŽ Inter-convertible:Inter-convertible:
ďŽ Odds = p/(1-p)Odds = p/(1-p)
Dr Max Mongelli 2016
34. Can tests be combined ?Can tests be combined ?
ďŽ Rare conditions: high rates ofRare conditions: high rates of
false positivesfalse positives
ďŽ Lead to excessive unnecessaryLead to excessive unnecessary
interventionintervention
ďŽ Can be reduced by combiningCan be reduced by combining
tests e.g. intrapartumtests e.g. intrapartum
monitoringmonitoring
Dr Max Mongelli 2016
35. Impact of new diagnosticImpact of new diagnostic
test on clinical outcomes:test on clinical outcomes:
ďŽ RCTRCT
ďŽ Cohort studyCohort study
ďŽ Case-control studyCase-control study
ďŽ Before and after studyBefore and after study
Dr Max Mongelli 2016
37. "It is surely a great criticism of our"It is surely a great criticism of our
profession that we have not organisedprofession that we have not organised
a critical summary, by specialty ora critical summary, by specialty or
subspecialty, adapted periodically, ofsubspecialty, adapted periodically, of
all relevant randomized controlledall relevant randomized controlled
trials."trials."
Archie Cochrane, 1972Archie Cochrane, 1972
Dr Max Mongelli 2016
38. Role of systematic reviewsRole of systematic reviews
ď§ Before commencing a new project: to determineBefore commencing a new project: to determine
whether further studies are reallywhether further studies are really indicated: âstate-indicated: âstate-
of-the-artâ literature review.of-the-artâ literature review.
ď§ Gain in statistical power for average estimates.Gain in statistical power for average estimates.
ď§ 'Cumulative' meta-analysis can determine when'Cumulative' meta-analysis can determine when
further studies are nofurther studies are no longer indicated.longer indicated.
ď§ Design of subsequent studies.Design of subsequent studies.
ď§ Setting policy for treatment and health care âSetting policy for treatment and health care â
making the best use of themaking the best use of the data available.data available.
Dr Max Mongelli 2016
39. Can Studies be Combined?Can Studies be Combined?
ď§ Identification of optimal inclusion criteria can beIdentification of optimal inclusion criteria can be
difficult.difficult.
ď§ The most critical step is choosing the appropriateThe most critical step is choosing the appropriate
research question.research question.
ď§ A fairly general question is more preferable to a veryA fairly general question is more preferable to a very
specific one.specific one.
ď§ Tukey : "...far better an approximate answer to theTukey : "...far better an approximate answer to the
right question, than an exact answer to the wrongright question, than an exact answer to the wrong
question.."question.."
Dr Max Mongelli 2016
40. Publication BiasPublication Bias
ď§ Entire research studies may fail to reach publicationEntire research studies may fail to reach publication
because of the nature of the results.because of the nature of the results.
ď§ Identification of unpublished trials can be veryIdentification of unpublished trials can be very
difficult - in one study it accounted for 22% of thedifficult - in one study it accounted for 22% of the
papers included in the meta-analysis.papers included in the meta-analysis.
ď§ Failure to publish rests with the investigators ratherFailure to publish rests with the investigators rather
than editors.than editors.
Dr Max Mongelli 2016
41. ď§ Comparison of the meta-analyses of smaller studiesComparison of the meta-analyses of smaller studies
with the corresponding result of the largest study.with the corresponding result of the largest study.
ď§ 30 meta-analyses including a total of 185 randomised30 meta-analyses including a total of 185 randomised
controlled studies (RCT) obtained from the Cochranecontrolled studies (RCT) obtained from the Cochrane
pregnancy and childbirth database. The meta-pregnancy and childbirth database. The meta-
analyses were only included if they had at least oneanalyses were only included if they had at least one
trial with a total sample size of over 1000.trial with a total sample size of over 1000.
ď§ Calculations differ from the Cochrane database inCalculations differ from the Cochrane database in
that the largest trial was excluded, this being used asthat the largest trial was excluded, this being used as
the 'gold standard' for outcomethe 'gold standard' for outcome
PREDICTIVE ABILITY OF META-ANALYSESPREDICTIVE ABILITY OF META-ANALYSES
Villar et al, Lancet 1995Villar et al, Lancet 1995
Dr Max Mongelli 2016
42. ďŽ There was total agreement between the meta-
analysis and the largest study in 18/30 (60%
[C.I. 42-78]) of comparisons.
ďŽ There was partial agreement between the
meta-analysis and the largest study in 6/30
(20% ) of comparisons.
ďŽ There was disagreement in 6/30 (20% [C.I. 6-
34] ) of comparisons.
PREDICTIVE ABILITY OF META-ANALYSESPREDICTIVE ABILITY OF META-ANALYSES
Villar et al, Lancet 1995Villar et al, Lancet 1995
Dr Max Mongelli 2016
44. The Cochrane CollaborationThe Cochrane Collaboration
⢠Established in 1993 by Sir Iain Chalmers
⢠International: 100 countries
⢠Independent
⢠Not-for-profit
⢠Over 27000 contributors
Dr Max Mongelli 2016
45. RANZCOG and EBMRANZCOG and EBM
âRANZCOG endorses the principles of Evidence-
based Medicine and recognizes the NHMRC levels
of evidence and grades of recommendationsâ
College Statement C-Gen 15, Nov. 2009
Dr Max Mongelli 2016
46. âŚa scientific idea can never be proven true,
because no matter how many observations seem to
agree with it, it may still be wrong. On the other
hand, a single contrary experiment can prove a
theory forever falseâŚ
Sir Karl Popper
Dr Max Mongelli 2016