The GATE Frame: a Generic Appraisal Tool for Epidemiology

1
The GATE Frame a Generic Appraisal Tool for
Epidemiology
Rod Jackson EPIQ July 2003
http//www.health.auckland.ac.nz/comhealth/epiq/ep
iq.htm
2
GATE
3
Evidence-based practice?
4
Evidence-based practice?
5
The 5 steps of EBP

• 1. translate information needs into answerable
  questions
• 2. search for the best evidence to answer them
• 3. appraise evidence for validity, impact
  applicability
• 4. integrate best evidence with
  patient/client/population values, professional
  judgment (expertise) costs apply in practice
• and
• 5. evaluate your performance (QI audit).

6
Clinical scenario

• 62 year old post menopausal woman with stable
  angina for 5 years.
• Visits you (her general practitioner) for a new
  prescription for her angina and HRT medications.
• Was started on HRT (as a cardio-protective
  medication) 3 years ago by a cardiologist who
  assess her angina.
• She has no menopausal symptoms, is a non
  smoker,BMI 26, blood pressure 140/80, total
  cholesterol 6 mmol/L, HDL 1 mmol/L, and a
  strong family history of CHD
• She has heard that HRT might increase the risk of
  CHD rather than prevent it and asks for your
  advice.

7
GATE approach design
Source Population
Who took part?
8
GATE approach design
Source Population
Eligible Population
Who took part?
9
GATE approach design
Source Population
Participants
Eligible Population
Who took part?
10
GATE approach design HERS
68,561 women screened from 20 outpatients/communit
y screening centres
source Population
Participants
Eligible Population
11
GATE approach design HERS
source Population
Participants
Eligible Population
Post-menopausal, established CHD, lt 80 yrs, no MI
in last 6 mths, no HRT last 3 months
12
GATE approach design HERS
All eligibles invited (2763)
source Population
Participants
Eligible Population
13
GATE approach design
Exposure Grp (intervention)
Participants
Comparison Grp (control)
What groups were compared?
14
GATE approachdesign HERS
HRT
what?
who?
Exposed (intervention)
Participants
Comparison (control)
Identical placebo
15
GATE approach design (multiple exposure groups)
EG1
EG2
Participants
EGn
CG
16
GATE approach design
EG
Participants
allocated?
CG
How were groups allocated?
17
GATE approach design (dichotomous outcomes)
Outcomes? -
EG
Participants
CG
What outcomes were assessed?
18
GATE approach HERS
Outcomes 1º CHD -
what?
who?
HRT
yes
no
Participants
Placebo
19
GATE approach design (multiple category outcomes)
Outcomes? 1 2 3 n
EG
Participants
CG
20
GATE approach design (continuous outcomes)
Outcomes? -
EG
high ? low
Participants
CG
21
GATE approach design HERS (continuous outcomes)
Outcomes HDL cholesterol -
what?
who?
Average 1.4mmol/L
HRT
Participants
Average 1.27mmol/L
Placebo
22
GATE approach design (longitudinal)
Outcomes? -
EG
Participants
CG
when? Time
When were outcomes assessed?
23
GATE approach design (cross-sectional)
Outcomes? -
EG
Participants
CG
when? Time
When were outcomes assessed?
24
GATE Frame 5 PECOT components
4.Outcomes -
2. Exposure Gp 3. Comparison Gp
1. Participants
5. Time
Every question you can answer from
epidemiological evidence has 5 parts (not 4)
because every study has these 5 parts
25
5 part question

• P - in post menopausal women with CHD
• E - does HRT
• C - compared with no HRT
• O - reduce the risk of further CHD events
• T - over the next 5-10 years

26
GATE Appraisal checklists

• Intervention studies
• Prognostic/Risk/frequency/X-section. studies
• Diagnostic test accuracy studies
• Case-control studies (Interv./Risk)
• Systematic review studies
• Download checklists from
• http//www.health.auckland.ac.nz/comhealth/epiq/ep
  iq.htm

27
GATE Appraisal checklists

• Study design numbers tables
• Internal study validity
• Study results (magnitude precision)
• External validity of study
• Download checklists from
• http//www.health.auckland.ac.nz/comhealth/epiq/ep
  iq.htm

28
GATE Appraisal checklists

• Section 1 Design numbers
• P
• E
• C
• O
• T
• See numbers later

29
GATE Appraisal checklists

• Section 2 Internal validity
• P
• E
• C
• O
• T
• Analyses

30
GATE Appraisal checklists

• Section 3 Study results
• Magnitude
• Precision
• Power

31
GATE Appraisal checklists

• Section 4 External validity
• P
• E
• C
• O
• T

32
Show checklist
show a checklist
checklist
33
GATE Frame validity
Outcomes -
measurement
EG CG
A B
Participants
Time
selection
confounding
measurement
34
Selection external validity issue
source Population
Outcomes -
EG CG
A B
Participants
Time
selection
eligible Population
35
GATE selection bias HERS
68,561 women screened from 20 outpatients/communit
y screening centres
All eligibles invited (2763)
source Population
Participants
Eligible Population
Post-menopausal, established CHD, lt 80 yrs, no MI
in last 6 mths, no HRT last 3 months
36
Confounding - present
Outcomes -
EG CG
A B
Time
confounding
37
Confounding - minimised by randomisation
Outcomes -
EG CG
A B
Random allocation
Time
confounding
38
Confounding - post randomisation Co-interventi
on
A B
EG CG
Time
confounding
other exposures
39
Measurement bias getting all the right
numbers in the right places
Outcomes -
Exposure Comparison
A C B D
EG CG
Time
measurement
measurement
40
Non-differential misclassification
Well defined? Replicable?
Outcomes -
Exposure Comparison
A C B D
EG CG
Time
measurement
measurement
41
Differential misclassification
Outcome assessed blind to intervention?
Exposure Comparison
Outcomes -
A C B D
EG CG
Time
measurement
Blind to intervention?
measurement
42
Diff. or non-diff. misclassification compliance
/ contamination
A B
EG CG
Time
measurement
43
Ascertainment Analysing All?
Outcomes -
EG1
EG2
A C B D
Allocated
Participants
CG2
CG1
measurement
44
GATE

• The numbers

45
EPIDEMIOLOGY

• The study of the OCCURRENCE of health outcomes in
  populations
• (over time)

46
OCCURRENCE

• Outcomes / Population / Time
• Numerator/Denominator/Time
• N
• D

47
(No Transcript)
48
GATE approach the numbers
NUMERATOR Outcomes -
DENOMINATOR
Exposed?
Participants
Comparison?
TIME
49
Occurrence

• Incidence
• measures outcome ONSETS (the number of onsets of
  outcomes in a population per time period)
• (N / D) /T a rate
• Prevalence
• measures outcome STATUS (the proportion of people
  in a population have had an outcomes). Point
  prevalence is assessed at one point in time.
  Period prevalence is assessed for a period of
  time.
• ( N / D proportion or percentage
• or ? (N) / D average or mean

50
Occurrence (background for teacher)

• Goal of epidemiology
• to measure the occurrence of outcomes in
  populations events/population over time per
  unit time
• there are two key measures of occurrence
  (incidence prevalence)
• Incidence
• measures outcome onsets (number of onsets of
  outcomes in a population per time period)
• (N / D) /T incidence rate or occurrence over
  time N/D during a period of T incidence
  proportion
• Prevalence
• measures outcome status (how many people in a
  population have had an outcomes) assessed at a
  point in time (point prevalence) or during a
  period of time (period prevalence)
• N / D a proportion or percentage or ?
  (N) / D an average or mean
• Prevalence depends on both incidence and the
  duration of the condition

51
Occurrence (Incidence) N / D / T
DENOMINATOR
NUMERATOR
Outcomes -
EG CG
A C B D
Participants
Population
TIME
52
Occurrence (Prevalence) N / D
DENOMINATOR
NUMERATOR
Outcomes -
EG CG
A C B D
Participants
Population
TIME 1
53
Mean Occurrence sum (N1 N2 Ni) / D
DENOMINATOR
NUMERATOR
Outcomes -
EG CG
N1 N2 ..Ni ..
Participants
Population
TIME 1
54
Occurrence

• Occurrence risk frequency
• incidence or prevalence
• EGO Exposure Group Occurrence A/EG/T
• CGO Comparison Group Occurrence
• B/CG/T
• See GATE calculator

55
GATE approach the numbers
Numerator Outcomes -
Denominator
EG1
EG2
A B
Allocated
Participants
CG2
CG1
56
GATE approach the numbers
Numerator Outcomes -
Denominator
EG1
EG2
A B
Allocated
Participants
CG2
CG1
EGO A/EG1 (Intention To Treat ITT analyses)
or EGO A/EG2 (On Treatment OT analyses)
57
Effect estimates

• Relative risk (RR) EGO
• CGO
• Risk difference (RD) EGO - CGO
• Number Needed to Treat (NNT)
• 1 / Risk Difference
• See GATE calculator

58
GATE approach HERS
Numerator Outcomes -
Denominator
EG11380
EG2
Allocated
Participants
CG2
CG1 1383
59
GATE approach HERS
Numerator Outcomes -
Denominator
EG11380
27
0
EG2
Allocated
Participants
CG2
0
32
CG1 1383
60
GATE approach HERS
Numerator 1CHD -
Denominator
EG11380
27
0
EG2
A172 B176
Allocated
Participants
CG2
0
32
CG1 1383
Time 4.1 yrs
61
Occurrence estimates

• EGO 172 CHD events / 1380 on HRT / 4.1 years
• 30.4 /1000 persons / year
• CGO 176 CHD events /1383 on placebo/ 4.1 yrs
• 31.04 / 1000 on placebo / year

62
Effect estimates

• Relative risk (RR) EGO
• CGO
• Risk difference (RD) EGO - CGO
• Number Needed to Treat (NNT)
• 1 / Risk Difference
• See GATE calculator

63
Effect estimates

• Relative risk (RR) EGO 30.40 0.98
• CGO 31.04
• Risk difference (RD) EGO - CGO
• 30.4/1000/yr - 31.04/1000/yr 0.64/1000/yr
• Number Needed to Treat (NNT)
• 1 / Risk Difference 1/0.64/1000/yr 1,562

64
GATE approach design HERS (continuous outcomes)
Outcomes HDL cholesterol -
what?
who?
Average 1.4mmol/L
HRT
Participants
Average 1.27mmol/L
Placebo
65
Occurrence estimates continuous data (e.g. HDL
cholesterol)

• EGO 1.40 mmol/L at end of year 1
• CGO 1.27 mmol/L at end of year 1

66
Effect estimates continuous data (e.g. HDL
cholesterol)

• Relative mean (RM) EGO 1.40 1.10
• CGO 1.27
• Mean difference (MD) EGO - CGO
• 1.40 mmol/L - 1.27 mmol/L 0.13 mmol/L
• Number Needed to Treat (NNT)

67
Influence of different presentation of
probabilities on clinical decisions
Risk Information presentation Accept Rx RRR A
cholesterol lowering 88 pill resulted in a
34 reduction in heart attacks ARR 2.5 had a
heart attack vs 42 3.9 - a difference of 1.4
NNT 71 patients treated for 5 yrs 31 to
prevent 1 heart attack
Hux Naylor Med Decis Making 199515152-7
68
the checklists
69
GATE intervention trials obs. studies
Intervention
Outcomes -
EG CG
A C B D
Participants
? randomised
Comparison
Time
70
GATE case series
Intervention
Outcomes -
EG CG
A C B D
Participants
not randomised
No Comparison
Time
71
GATE intervention before/after studies
Intervention
Outcomes -
EG CG
EG CG
A C B D
Participants
? randomised
Comparison
Time
72
GATE non-randomised incidence studies of
prognosis/prediction/risk/frequency
Risk factor
Outcomes -
EG CG
A C B D
Participants
measured
Comparison
Time
73
GATE prevalence studies
Risk factor
Outcomes -
EG CG
A C B D
Participants
measured
Comparison
Time
74
How good is the test?
75
GATE diagnostic test accuracy studies
Reference standard ve
Test -
EG CG
A C B D
Participants
measured
Reference standard - ve
Time
76
GATE diagnostic test accuracy studies
Reference standard ve
Test -
EG CG
A C B D
Participants
measured
Reference standard - ve
Time
77
GATE case-control studies of intervention / risk
Risk factor
Outcomes -
EG eg CG cg
A c C B d D
Population
measured
Comparison
Time
78
Alternative GATE
Participants
Exposure Group Comparison Group
EG
CG

-
OutcomeTime
B
A

D
C
-
79
Is the effect precise?

• Absence of random error
• (a new presentation is under development)

80
GATE Frame random error
4.Outcomes -
1. Participants
2. Exposed 3. Comparison
5. Time
Incidence/ prevalence/ RR / RD / NNT (
confidence interval, p-value)
81
Ken Rothman/s episheet www.oup-usa.org/sc/0195135
547/downloads.html An excellent free excel
spreadsheet for calculating confidence intervals
and power
82
Power
high
low
83
Power
high
low
84
Power
Very high
High enough
85
Are the findings applicable?

• relevant, feasible, affordable, generalisable

86
Applicability Participants
source Population
Outcomes -
EG CG
A B
Participants
Time
eligible Population
87
Applicability Exposure Comparison
source Population
Outcomes -
EG CG
A B
Participants
Time
eligible Population
88
Applicability Outcomes Time
source Population
Outcomes -
EG CG
A B
Participants
Time
eligible Population
89
(No Transcript)
90
?
?