Evidence Based Medicine: Review of the basics

Report
Evidence Based Medicine:
Review of the basics
DEB BYNUM, MD
AUGUST 2010
Moving beyond sensitivity and specificity
Hierarchy of Strength of Evidence
 N of 1 randomized controlled trial
 Systematic reviews of randomized controlled trials
 Single randomized trial
 Systematic review of observational studies
addressing patient important outcomes
 Single observational study addressing patient
important outcomes
 Physiologic studies
 Unsystematic clinical observations
Where to start?
 The Clinical Question
 Focus

Question types:
Diagnosis
 Harm
 Prognosis
 Treatment

Next Step: Research
Basic Steps in reviewing an article
 1. Are the results of the study valid?
 2. What are the results?
 3. How can I apply these results to patient care?
Therapy
 1. Are the results valid?
 Were patients randomized?
 Was randomization concealed?
 Were patients analyzed in the groups to which they were randomized?
 Were patients in treatment and control groups similar?
 Were patients aware of group allocation? (?blinded?)
 Were clinicians aware of group allocation?
 Was follow up complete?
 2. What are the results?
 How large was the treatment effect?
 How precise was the estimate of treatment effect?
 3. How can I apply the results to patient care?
 Were the study patients similar to the patient in my practice?
 Were all clinically important outcomes considered?
 Are the likely treatment benefits worth the potential harm and costs?
Therapy: Analyzing Results
 How good was the treatment?

Absolute Risk Reduction

Relative Risk

Relative Risk Reduction
Absolute Risk Reduction
 Absolute difference in outcomes between the
treatment and control groups

X% have outcome in control group, Y% have outcome in
treatment group

ARR = X-Y
Relative Risk
 Risk of events among patients in the treatment group
compared to/relative to the risk in the control group

Y/X
Relative Risk Reduction
 RRR
 (1-Y/X) x 100
Why is this the most commonly reported measure of
treatment effect?
How can this be misleading????
RRR … small treatment effect
 Pink Potion is new on the market, and is a wonder
drug according to Big Bucks Pharmaceutical
Company. When taken every day for 20 years, it
decreases the risk of developing cell phone related
brain cancer by 50% ! (RRR is 50%)
 You find the risk of such cancers to be 0.25 % in one
high risk population. With this 20 year treatment,
0.13% of the population developed the cancer…
Risk Reduction: Take Home
 Beware RRR
 Calculate ARR whenever possible – this often will
put the overall treatment effect into perspective
 Most reports (especially when
advertising/promoting a treatment) will report
effects as RRR…
How precise was treatment effect?
 Confidence Intervals: 95%

95% CI defines a range of results that includes the true treatment
effect result 95% of the time
 P <.05
 Larger sample size – narrower the CI
 Using CI to estimate treatment importance


“positive study”– look at lower limit for CI, if that number is the true
treatment effect, is that important/beneficial when applied to your
patient?
“negative study” –look at upper limit for CI– if that number is true,
would that be clinically important? Study may not prove that a
treatment is of benefit, but at same time may not prove that it is not
of benefit…
Treatment: Applying results to patient care
 Were the study patients similar to patients in my
practice?
 Were all clinically relevant outcomes considered?
 Are the likely treatment benefits worth the potential
harm and costs?

Number Needed to Treat (NNT)
NNT
 Number of patients who must receive a
treatment/intervention to prevent one bad outcome
or produce one positive outcome
 1/ ARR
 (Pink Potion: NNT = 1/.12% = 1/.0012= 833== need
to treat 833 people for 20 years to prevent one case
of cell phone related brain cancer…)
Number Needed to Harm (NNH)
 Absolute risk of adverse outcome with treatment –
risk of adverse outcome without treatment
 NNH = 1/ absolute difference in adverse outcomes
(just like NNT)
 Weight NNT with NNH….
Reviewing studies looking at Harm
 Are the results valid?
 Was there demonstrated similarity in all known determinants of
outcome? Did they adjust for differences in their analysis?
 Were exposed patients equally likely to be identified in the two groups?
 Were outcomes measured the same way each group?
 Was follow up sufficiently complete?
 What are the results?
 How strong is the association between exposure and outcome?
 How precise is the estimate of risk?
 How can I apply the results to patient care?
 Were the study patients similar to my patients?
 Was the duration of follow up adequate?
 Was was the magnitude of the risk?
 Should I try to stop the exposure?
Design
Starting
Point
Assessment
Strengths
Weaknesses
Cohort
Exposure
status
Outcome
event status
Feasible when
not able to
randomize
Bias, limited
validity
Case-control
Outcome
event status
Exposure
status
Overcomes
delays, may
only need
small sample
size
Bias, limited
validity
RCT
Exposure
status
Adverse event
status
Low
susceptibility
to bias
Feasibility,
generalizabilit
y
Assessing Harm in Case Control Studies
 Cannot use RR (RR depends upon determining the
proportion of patients with outcome – in a case
control study, the proportion of individuals with an
outcome is chosen by the investigator)
 Odds Ratio:
 Odds of a case patient being exposed / odds of a control
patient being exposed
OR and RR
Outcome: yes
Outcome: no
Exposure: yes
A
B
Exposure: no
C
D
RR= a/(a+b) / c/(c+d)
OR = (a/b) / (c/d)
If outcome is infrequent in both treatment and control groups, then OR and RR
Be nearly the same
Studies looking at Harm
 Can I apply the results to my patient?

Were the study patients similar?

Was the duration long enough?

How big is the risk?

Should I stop the exposure? (does risk outweigh any benefit?
Is there alternative therapy with less risk?)
Diagnostic Testing
Thresholds
 Probability of diagnosis: 0-100%
 Test threshold
 Treatment threshold
 Probability below test threshold: no testing
 Probability in between : test
 Probability above treatment threshold: no testing,
treat
Looking at Diagnostic Tests
 Are the results valid?
 Did clinicians face diagnostic uncertainty
 Was there a blind comparison with an independent Gold Standard
applied to both Treatment and Control groups
 Did the results of the test being studied influence the decision to perform
the reference/gold standard?
 What are the results?
 What are the Likelihood Ratios associated with the range of possible test
results
 How can I apply the results to patient care?
 Will the reproducibility of the test result and its interpretation be doable
in my clinical setting?
 Are the results applicable to my patients?
 Will the results change my management strategy?
 Will patients be better off as a result of the test?
Sensitivity and Specificity
Disease +
Disease -
Test +
A
B
Test -
C
D
Sensitivity: If the patient has the disease, how likely is it that he will
have a + test?
a/a+c
“rules out” disease (not really….)
Specificity:
If the patient does not have the disease, how likely is he to have
a negative test?
d/b+d
“rules in” disease (again, not really….)
The problem with sensitivity and specificity
 In real life, the question is “the patient has a positive
test, how likely is it that he has the disease? Or the
patient has a negative test, how likely is it that he
does not have the disease?”
Predictive Values
Disease +
Disease -
Test +
A
B
Test -
C
D
Positive Predictive Value:
How many (%) people with a positive test will have the disease?
PPV = a/ a+b
Negative Predictive Value:
How many people with a negative test will NOT have the disease?
NPV = d/c+d
Problem: depend upon prevalence (low prevalence population, positive test is more
Likely to be a false positive; high prevalence, a negative more likely to be a false
negative)
Likelihood Ratios
 LR does not depend upon prevalence and can be more
easily applied to a specific patient with a known test
result to estimate the post-test probability that the
patient has “disease”
 Points:




Not affected by prevalence
Can be made specific to your patient (based upon pretest probability
of disease)
Can be linked with other LRs to come up with a post test probability
Does not rely upon test being dichotomous (positive or negative)
Likelihood Ratio
 The likelihood that disease is present given X test
result (positive, negative, intermediate, 250)
 LR: How many patients with X test result HAVE
disease compared to number of patients with X test
result who do NOT have disease
Points about the LR
 X test result can be positive, negative, intermediate,
a number
 LR always looks at ONE certain test result and
always compares likelihood of having disease to
likelihood of not having disease
LR for a positive test (Positive LR)
 Likelihood that disease is present given a “positive” test
 How many patients with a + test HAVE disease





compared to # patients with a + test who do NOT HAVE
disease
True Positive Rate/False Positive Rate
LR + test: a /(a+c) / b/(b+d)
= sensitivity/1-specificity
Higher # (over 10) : better predicting
LR + “infinity” = 100% specificity (if the test is positive,
the patient has disease, no false positives)
LR for Negative Test (-LR)
 Likelihood that disease is present given a negative test
 How many patients with a Negative test HAVE disease compared to
# patients with a Negative Test who DO NOT HAVE disease
 False negative rate/true negative rate
 c/(a+c) / d/(b+d)
 1-sensitivity/ specificity
 Smaller number = better test (fewer patients with a negative test
HAVE disease compared to DO NOT HAVE disease)
 LR – of <.10 usually signficant
 LR – of 0 = 100% sensitivity (if the test is negative, the patient does
NOT have disease – rules “out” disease…
Review the chart again…
Disease +
Disease -
Test +
A
B
Test -
C
D
Sensitivity: a/a+c
Specificity: d/b+d
PPV=a/a+b
NPV=d/c+d
LR + test: a/(a+c) / b/(b+d)
LR – test: c/(a+c) / d/(b+d)
How to use the LR
 LR is ODDS note a %
 Determine pretest probability (ok to estimate)
 Determine pre test ODDS (odds= probability/1-
probability)
 Determine Post test ODDS: pretest odds x LR
 Convert post test ODDS back to probability
LR Nomogram
 P. 129
Looking at Summaries of Evidence
 Systematic Reviews
 Meta-analyses
Summary articles
 Are the results valid?
 Did the review explicitly address a sensible clinical question?
 Was the search for relevant studies detailed and exhaustive (publication
bias)
 Were the primary studies high quality?
 Were assessments of studies reproducible?
 What are the results?
 Were the results similar from study to study?
 Were the outcomes the same (comparing apples to apples…)
 What are the overall results?
 How precise were the results?
 How can I apply the results to patient care?
 How can I interpret the results?
 Were all clinically important outcomes considered?
 Are the benefits worth the costs and potential risks?
Developing a CAT sheet:
 Critically Appraised Topic
 Clinical Question
 Clinical Bottom Lines
 Methods
 Summary of Results (create table if needed)
 Comments (strengths, weaknesses, limitations)
 How can I apply this to my patients?
 References

similar documents