Transcript Sample size

Study Designs-Review
Epidemiology 200A
Dr. Jørn Olsen
November 24, 2009
Fall 2009
1
Causation
1st Axiom: Diseases exist as a function of preceding causes
causes exist
causes (some) have a social or environmental history
2nd Axiom: Diseases are distributed as a function of
susceptibility and exposures
Host
Exposure
Environment
Diseases follow sufficient causal fields
2
Causation
3rd Axiom: The manifestation of diseases is a function of
treatment and natural defense mechanism.
4th Axiom: Diseases are not distributed at random. They are
not unavoidable events.
Some causes are man made.
Therefore, some diseases are preventable.
3
Hume 1711-1776:
We may define a cause to be an object followed by
another and where all the objects similar to the first are
followed by objects similar to the second.
E
D
4
Not many examples of E D associations in medicine.
Causes in the ”strong-Hume” sense could be seen as
necessary and sufficient causes
A necessary cause
E
D
several examples in medicine, but they are usually
man-made
tuberculosis
post-partum depression
AIDS, etc.
A sufficient cause
E
D
5
Sexual debut is a risk indicator, maybe a risk factor or a
cause (sexual activity is a necessary cause).
E
D
6
Causes act conditionally upon other causes and in
combination they become a sufficient causal field.
Each component cause is only sufficient given the other
component causes exist.
I
A
II
B
A
III
D
C
70%
B
E
F
20 %
10%
Causes are only necessary for a subdomain of diseases (I,
II or III) and they are only sufficient given the other
causes are in place.
The strength of a given association depends upon the
7
prevalence of the other component causes.
DAG
A
air p
B
sex
C
bronchial
response
treatment
E
arc
asthma
D
A
A-C-D directed (causal)
B adjacent
> nodes
E-C-D not directed
Ancestor – descendant
Acyclic if no directed path forms a closed loop
8
Measures of disease occurrence, measure of
associations
9
Breast cancer screening trial 5year follow-up
Screening
N
Obs. Time
Death
Yes
25,000
118,000
180
Yes
25,000
117,500
210
CI+, R+
180/25,000 = 0.0072
CI-, R-
210/25,000 = 0.0084
IR+
180/118,000 = 0.00153
IR-
210/117,500 = 0.00179
10
If disease is rare…
R~ IR ~ disease-odds=
(180/25,000) ~ (180/118,000) x 5 ~ (180/25,000) / (24820/25,000)
0.0072 ~ = 0.0076 ~ 0.0073
11

From incidence rates to incidence proportions Kaplan-Meir or
exponential formula:
∑IR Δt
k
k
k
Conditions:
1-e
1)
Closed population
2)
No competing risks
3)
Stable IR over Δt and IR is small
12
Associations
RR = R+ / R- = 0.857
IRR = IR+ / IR- = 0.853
OR = Disease odds + / Disease odds - = 0.857
13
Attributable fractions
 Screening a preventive measure (for mortality). Treat
lack of screening as ‘a causal exposure’.
 210 die in the non-screened group. Had they had the
same risk as the screened we would expect
(180/25,000) x 25,000 = 180 deaths
 30 deaths prevented out of 210 = 30/210 x 100 = 14%
14
Population
or ( R- - R+) / R- = 1-RR
(0.0084-0.0072) /0.0084 = 1-0.857 =0.14
 Preventive fraction-Vaccine efficiency in infectious
disease epidemiology
15
 210+180 died in the population. 30 could have been
prevented = 30/390 = 7.7%
 Formal procedure
R- x N- - R+ N -
R- x N- + R + x N+
16
17
18
Prevalence and incidence
Ix D = P/(N – P) in steady state population with no migration.
I(N-P)Δt
I (N-P) Δt = (1/D) P Δt
I (N-P) = 1/D P
I1PΔt or
(1/D)PΔt
I1 = incidence rate of
leaving the
prevalence pool
I x D = P/(N-P)
P/(N-P) prevalence odds; if P is small
I x D ≈ P/N
19
 Is the obesity epidemic leading to an epidemic of
diabetes?
 The prevalence of diabetes is increasing but ?
20
21
22
Randomized Controlled Trial (RCT)
 Experimental studies: The researcher manipulates the
exposure in order to study the effect of the exposure.
 The unit of observation is often individuals, but can be
regions.
 Requires longitudinal data collection.
 Relies upon primary data or a combination of primary and
secondary data.
 Is well suited to estimate the expected therapeutic efficacy in
quantitative terms under ideal circumstances (efficacy). Does
not usually provide evidence of unexpected effects (or side
effects). Does not provide evidence of effects in routine
circumstances (effectiveness).
23
Three key elements in a classic RCT
Randomization: comparability of populations
Placebo:
comparability of circumstances
Blinding:
comparability of information
24
Basic structure of the RCT
 Set up a specific hypothesis
 Select a sampling frame among suitable patients
 Define inclusion/exclusion criteria
 Obtain informed consent
 Randomize
 Follow compliance
 Measure outcome
 Analyse according to intention to treat or
according to protocol
25
Sequence
Candidate patients
Define source
population / study base
incl./excl. Criteria
Informed consent
randomize
Compliance
Loss to follow-up
Effect
Analyse:
Intention to treat
or
According to protocol
26
Reporting from the trial should at least provide the
information requested by the CONSORT group (Lancet
2001; 357: 1191-94).
CONSORT: Consolidated Standards of Reporting Trials
www.consort-statement.org
22-item checklist
Detailed explanation of CONSORT in Altman et al.
Annals of Internal Medicine 2001; 134: 663-694.
27
Screening for colorectal cancer
Kronborg et al. Lancet 1996; 348: 1467-71.
Figure 1: Study profile
28
Screening for colorectal cancer
Kronborg et al. Lancet 1996; 348: 1467-71.
Table 2. Compliance during repeated screening
Screening
round
Number of people
invited for screening
Number of people
screened
1
2
3
4
5
30,762
20,113
18,236
16,746
15,279
20,672 (67%)
18,781 (93%)
17,279 (94%)
15,845 (94%)
14,203 (92%)
29
Screening for colorectal cancer
Kronborg et al. Lancet 1996; 348: 1467-71.
30
Classification of designs
Unit of observation
= individual or clusters of individual
macroepidemiology / microepidemiology
Allocation of exposure
= experimental / non-experimental
Longitudinal recording
= survey / case-control or follow-up
Selection of exposure
or outcome
= follow-up / case-control
Source population
= case-control; primary, secondary
Data
= primary / secondary
Data collection
= prospectively / retrospectively
31
31
Source: Kunzli et al. The Semi-individual Study in Air Pollution Epidemiology; A Valid Design as Compared to Ecologic
32
Studies. EHP 1997, 105 (10).