1
Q
describe an RCT
A
study where group of eligible people gets randomly assigned by an investigator to either an invention or control condition
2
Q
what is a group of eligible people called?
A
study sample
3
Q
intervention efficacy is evaluated by
A
comparing outcomes among those receiving intervention vs. control therapy/intervention
4
Q
key characteristics of RCTs
A
randomization
blinding
control/placebo vs “controlled” trial
5
Q
randomization is comprised of two types of (x)
A
allocation
6
Q
random allocation example
A
random # table
computer generated programs
sealed enveloped w/ randomization info
7
Q
nonrandom allocation
A
alternate assignment of treatments
assignment by day of the week
8
Q
random sample ensures
A
GENERALIZABILITY of survey results
9
Q
randomization ensures
A
COMPARABILITY of experimental group vs control group when participant pool is LARGE
10
Q
when is randomization unethical?
A
an effective treatment already exists
when personal choice is involved
risks of new treatment likely to exceed risks of existing treatment
11
Q
equipoise is when
A
there MUST BE GENUINE doubt about efficacy of treatment but SUFFICIENT belief it may work
12
Q
stopping rules for RCTs:
A
Beneficial (and should not be withheld from placebo group)
Harmful (and trial should be stopped)
Evidence is inconclusive (and trial should continue)
13
Q
Analyzing by intention to treat states that all participants must be
A
analyzed based on original assignment otherwise randomization is broken and groups aren’t comparable
14
Q
blinding is used to avoid bias in
A
Enrollment
During trial
Follow-up
15
Q
levels of blinding
A
single blind
double blind
triple blind
16
Q
single blind is when
A
participants are blinded but investigators are aware of intervention and control arm
17
Q
double blind is when
A
participants + investigators are not aware who’s receiving intervention
18
Q
triple blind is when
A
participants and investigators don’t know INTERVENTION assignment
Data analyses are done in a way that investigators are removed
19
Q
“control” condition provides
A
comparison arm by which investigator can compare the effect of treatment
20
Q
controls may receive NO treatment if there’s
A
no standard of care (unethical otherwise)
21
Q
types of RCTs
A
natural experiments
community trials
cluster randomized trials
individual level randomization
22
Q
example of a natural experiment
A
john snow - cholera and the broad street pump
“resembles a planned trial”
23
Q
community trials example
A
water fluoridation trials
where one group gets an intervention and the other doesn’t
24
Q
cluster randomized trials example
A
flu vaccines being disseminated in some communities and not others to assess herd immunity
CLUSTERS!!!!
25
individual level randomization randomizes
eligible individuals to an intervention (TREATMENT) or a control/placebo/standard of care condition
26
advantages of RCTs
demonstrates cause-effect relationships
may be faster and cheaper than cohort studies
allows investigators to control exposure levels as needed
27
disadvantages to RCTs
only ethically appropriate for SOME research questions
more resource intensive
many interventions not suitable for blinding
tested interventions may be diff than common practice
limited generalizability due to the use of volunteers, eligibility criteria, loss to f/u
28
RCT measures of association
odds ratio (outcome at fixed pts in time)
incidence rate ratio (events measured in person-time)
hazards ratio (time to event)
29
prevalence for comparing disease among exposed vs. unexposed:
a/[a+b] vs. c/[c+d]
30
prevalence ratio for comparing diseased vs. not diseased:
a/[a+c] vs b/[b+d]
31
measures of association for cross sectional studies
prevalence ratios
prevalence odds ratios
32
strengths of cross sectional studies
easy to conduct
cost-effective
yields basic associations
33
weaknesses of cross sectional studies
Identifying prevalence (P = ID)
Length based bias (cases that are around longer are more likely to be picked up)
can’t establish causality
34
measures of association for cohort studies
risk ratio
rate ratio
OR
35
cohort study strengths
Select groups based on exposed/unexposed status
Recall bias minimized
Can study multiple exposures
Minimize bias in ascertainment of exposures and covariates; esp. In prospective studies
Can study multiple outcomes**
Time order established
Efficient for rare exposures
Provides incidence rate of disease
36
cohort study weaknesses
Loss to follow-up
If large number of subjects is required or long follow-up = $$ or logistically challenging; especially for prospective design
Hard to study rare diseases
Changes over time in staff/methods
Little control over nature and quality of data in retrospective designs
37
cumulative incidence ratio (CIR) aka
risk ratio
38
risk in exposed =
a/(a+b)
39
risk in unexposed =
c/(c+d)
40
risk ratio =
a/(a+b) / c/(c+d)
41
incidence density ratio (IDR) aka
rate ratio
42
rate in exposed =
a/PY exposed
43
rate in unexposed =
c / PY unexposed
44
rate ratio =
a/PY exposed / c/PY unexposed
45
prospective cohort studies pros/cons
pros: better exposure data, less bias
cons: expensive time, consuming
46
retrospective cohort studies pros/con
pros: cheaper, faster, effective for diseases w/ long latency
cons: exposure data limited
47
measure of association for case-control
exposure odds ratio
48
odds of exposure among cases
a/c
49
odds of exposure among controls
b/d
50
exposure odds ratio =
ac/bd
51
Ratio for comparing DISEASE among exposed vs. unexposed =
PRd = [a/(a+b)] / [c/(c+d)]
52
Ratio for comparing EXPOSURE among diseased vs no disease =
PRe = [a/(a+c)] / [b/(b+d)]`
53
random error comes from
1) random processes
2) sample size
54
random processes is the random variation in...
sampling methods, data collection
interpretation causing results to change unpredictably
55
as sample size increases,
likelihood of random error decreases
56
systematic bias stems from
design, conduct or analysis of a study
57
systematic bias results in
mistaken estimate of an exposure’s effect on disease
58
information bias aka
measurement bias
59
information bias is always a...
a threat --> Cases may be misclassified as unexposed/exposed and vice versa
60
potential sources of information bias
Respondent
Data collector
Data managers
Data analysts
Study investigator
61
common types of info bias
Bias from surrogate interviews (spouses/friends/etc)
Surveillance bias (aka detection bias)
Recall bias
Reporting bias → wish bias
62
reduce info bias with
Precise definitions
Detailed measurement protocols
Repeated measurement
Proper training/certification of study staff
Data audits
Data cleans
Re-running all analyses prior to publishing
63
‘differential’ misclassification of exposure/outcome means
information errors occur differently for groups being compared
64
measures of association for differential misclassification
true OR
biased OR
65
differential misclassification bias (neg or pos?)
positive (AWAY from null)
66
‘non-differential’ misclassification of exposure/outcome means
information errors don't occur differently for groups being compared
67
bias for non-differential misclassification
NEGATIVE (TOWARD null)
68
missing data in observational studies
cohort study - exposure is out of investigator's control
69
positive bias direction
AWAY from null
70
for positive bias, observed value is
greater than true value
71
negative bias direction
TOWARD null
72
negative bias's observed value is
smaller than true value
73
what is a confounding factor/confounder?
estimate of effect (MoA) of exposure on outcome is distorted
74
result of confounding =
distortion of true MoA toward null (neg confounding) or away from null (pos confounding)
75
neg confounding direction
towards null
76
pos confounding direction
away from null
77
traditional approach to IDing confounder
- known risk factor for outcome
- associated w/ exposure
- not a result of exposure
78
when there's a non-causal association b/w an exposure and outcome that means...
there's a confounding variable / confounder
79
confounder is NOT an error but..
reflects an underlying true association b/w other variables with exposure and outcome
80
crude OR is larger or smaller than adjusted OR with positive confounding?
LARGER
81
crude OR is larger or smaller than adjusted OR with negative confounding?
SMALLER
82
How to control for confounding during the ANALYSIS phase of different epi studies
stratification
adjustment
83
How to control for confounding during the DESIGN phase of different epi studies
matching
randomization
restriction
84
stratification allows investigator to
hold confounder of interest 'constant' within a strata of the exposure
85
what is a strata
category
86
limitations of stratification
as # of confounders increase, size of each stratum gets very small
87
adjustment is when you use...
Use stat techniques to estimate what the association would be if the confounder was NOT associated w/ the exposure
i.e. Multivariable regression analysis
88
Multivariable regression analysis can be used to
Adjust for confounder through stat modeling
Can adjust for multiple confounders at the same time
89
matching to minimize confounding during the design phase is when you only have...
a few variables to match on
90
disadvantages of matching
Logistically difficult
Do not want to overmatch
Can’t match on all characteristics
Can’t study those factors you’ve matched on
91
randomization studies "assign" people....
at random to treatment vs. control condition
Randomization is considered to make treatment groups similar
92
randomization studies ensure...
comparability of the experiment group and control group when pool of study participants is large
93
what is created by randomization studies during the design phase of studies to reduce confounding
Exchangebility
94
restriction limits...
subject population to those within specified categories of extraneous factors
95
effect measure modification (EMM) occurs when...
relationship between exposure and outcome is DIFFERENT across different levels of a 3rd factor
96
To understand effect modification we must...
specify what’s being modified
97
Crude estimate is NOT used to evaluate...
the presence of interaction for EMM
98
to evaluation a rxn for EMM, you must...
Compare stratum-specific estimates directly to look for differences across exposure levels
allows you assess differences in EM estimates
99
confounding depends on...
distribution of confounder among strata of interest’s exposure
100
confounder obscures...
true relationship b/w an exposure and outcome
101
NUISANCE EFFECT is TO BE
ADJUSTED FOR
102
EMM is an inherent feature of...
the strata to be described
103
EMM alters...
effect in size/direction among strata
104
MoA for EMM
risk ratio
rate ratio
OR
105
risk ratio aka relative risk
a/(a+b) / c/(c+d)
106
rate ratio aka incidence rate ratio
a/PYexposed
c/PYunexposed
107
what does causal interference mean?
if an association is present, we have to determine if the exposure is truly the cause
108
temporality states
exposure must precede disease
109
strength of association is reflected in...
measure of association value
larger assocation --> more likely exposure is causing disease
caveat: weak associations may be causal but harder to rule out bias/confounding
110
biological gradient aka
dose-response relationsips
111
biologic gradient is the changes in
levels of exposure and how they're related to changes in risk of disease
caveat: thresholds/saturation points may exist but there will be no outcomes before/past a certain level of exposure
112
cessation of exposure
risk of disease goes down if potential cause is removed
caveat: if disease has already started, removal of cause doesn't reduce risk
113
replication of findings is the relationship b/w...
exposure and outcome
more likely to be causal
caveat: sometimes there are good reasons why study results differ
114
why is good that sometimes study results differ?
heterogeneity of effect across social/cultural settings
systematic review w/ meta analysis
115
coherence with established facts
If a relation if causal, one would expect observed findings to be consistent w/ other epidemiologic and biologic knowledge
116
caveat of coherence w/ established facts
data may not be available to reinterpret existing understanding of disease process in the face of new evidence
117
biological plausibility
Proposed mechanism should be etiologically plausible
Caveat: problematic for new types of causes
118
Consideration of alternate explanations is the extent to which...
investigator has ruled out other possible explanations
Caveat: alternative explanations limited by understanding of biology and sophistication of analysis
119
Specificity of association is associated with...
only one disease
Caveat: exposures are linked to MULTIPLE disease
120
A cause of a specific disease is an...
ANTECEDENT event
121
Rothman's causal inference is the condition/characteristic necessary for the disease when....
it occurred
and without which the disease event wouldn’t have happened
GIVEN THAT OTHER CONDITIONS ARE FIXED
122
Disease processes tend to be...
MULTIFACTORIAL
123
multicausalty means that...
Disease processes tend to be MULTIFACTORIAL
124
VERY FEW EXPOSURES cause...
disease entirely by themselves
125
examples of multicausality
E.g. exposure to measles can cause measles only if somebody is susceptible (e.g. not vaccinated)
Development of melanoma is among those w/ high UV light exposure who also have fair skin
126
steps in outbreak investigation
1) verify dx
2) confirm outbreak status
3) case definition
4) descriptive epi
5) develop hypothesis
6) test hypothesis
7) refine hypothesis + conduct additional studies
8) implement control + prevention measures
9) communicate findings
127
main measures of association for outbreak investigation
case control --> exposure OR
cohort --> risk ratio
128
verify diagnosis by asking:
1) is it a known agent
2) what do we know about transmission
3) ID and dx?
129
confirm outbreak status
real vs artifactual causes
130
case definition is the standard set of...
criteria for deciding whether an individual should be classified as having the disease of interest
131
advantages of case definition
1) Lab confirmation increases specificity of case definition
2) Increased specificity = increased likelihood of correctly IDing true negative persons
3) Reduces misclassifications = less likely to include as cases b/c negatives are being identified
4) Maxes power to detect source
132
disadvantages of case definition
1) Lab confirmation excludes patients who didn’t seek care, were not tested or tested without PFGE
2) Decreases sensitivity of case definition
3) Limited cases
4) Are dates reasonable
133
what does descriptive epidemiology mean?
characterizing cases
134
what makes up an epidemic curve?
mode of transmission
timing of exposure
course of exposure
135
epidemic curve sources
common source or propagated source
136
types of common source
point exposure
intermittent exposure
137
types of propagated sources
single exposure, no secondary cases
secondary and tertiary cases
138
point exposure can help...
ID incubation period for NEW agents
139
intermittent exposure can help identify...
time of exposure for KNOWN agents
140
example of single exposure, no secondary cases (propagated source)
measles
141
example of secondary and tertiary cases (propagated source)
hep A
142
how to develop a hypothesis
Conduct surveys or interviews
Hypothesis-generating interviews to explore all potential sources of infection w/ limited # of patients
Detailed surveys to obtain info on sociodemographic info, clinical details of disease, complete exposure history for period prior to onset
143
MoA for cohort study to test hypothesis
attack rate ratio aka risk ratio
144
primary attack rate =
of people at risk who develop the disease / total # of people at risk
145
primary attack rates are calculated among persons who...
acquire disease directly associated w/ an exposure
146
food-specific attack rate =
of people who ate specific food & develop illness / total # of people who ate the food
147
secondary attack rate =
total # of cases - initial case(s) / # of susceptible persons in group - initial case(s)
148
secondary attack rates are calculated among persons who...
acquire disease from exposure to primary case
measures infectivity of gent + effects of prophylactic agents
149
attack rate for exposed =
a/(a+b)
150
attack rate for unexposed =
c/(c+d)
151
attack rate ratio aka risk ratio =
(a/a+b) / (c/c+d)
152
MoA for case control
exposure OR
153
refine hypothesis step
What control measures to consider
What further studies can be done
Traceback study + results
154
implement control + prevention measures
1) Immediate problem
2) Larger issue
155
last step of outbreak investigation
communicate findings
156
what is a screening?
Presumptive ID of unrecognized disease or defect through tests/exams/rapid procedures
Not intended to be diagnostic
157
levels of prevention
Primary Prevention
Secondary Prevention
Tertiary Prevention
158
screening requirements
suitable disease
suitable test
suitable screening program
159
test characteristics
reliability
validity
160
reliability is the ability of a test to...
have consistent results
161
variation can stem from
Variation b/w individual subjects
Variation b/w those reading/interpreting test results
162
validity is correctly classifying...
people with pre-clinical disease as positive and those without pre-clinical disease as negative
163
validity measures
sensitivity
specificity
164
sensitivity is the probability of
correctly classifying those with DISEASE as CASES
165
specificity is the probability of...
correctly classifying those WITHOUT DISEASE as NON-CASES
166
what is a binary outcome
yes or no
167
discrete/continuous outcomes comprise of
categorical outcomes
scales
continuous outcomes (height, weight, BMI, etc)
168
what does a low-value cutoff mean
no false negatives (high SN)
many false positives (low SP)
169
what does a high-value cutoff mean
no false positive (high SP)
many false negatives (low SN)
170
what does a mid-value cutoff mean
minimizes false positive --> INTERMEDIATE SP
minimizes false negatives --> INTERMEDIATE SN
171
positive predictive value (PPV) =
true positives / all positives
a / a + b
172
PPV is a function of
disease prevalence + test sensitivity
173
negative predictive value (NPV) =
true negatives / all negatives
d / c + d
174
individual-level screening
Occurs at individual patient-doctor level (annual check up)
aka case finding → BP screening, PSA screening
Focus is on IDing existing disease in patient who don’t know they have it
175
population-level screenings
National level policy decision to offer mass screening to a whole sub group of a population
E.g. vision and hearing screening for all NYC children
176
how to evaluate screening programs
acceptability measures
efficacy of screening
efficiency
balance of risks (harms) vs benefits
bias
177
acceptability measures comprise of
convenience
safety
costs
subsequent fx testing
178
efficacy of screening is the potential to
reduce further morbidity and mortality
179
RCTs often employed to evaluate the...
efficacy of screening test
180
aspects of RCTs
intervention arm
control arm
outcomes
limitations
181
are RCTs or case-control/cohort better for testing screening efficacy?
RCTs
182
efficiency of screening programs is checked by
PPV
risks + costs of follow up of those testing positive
cost-effectiveness
183
compliance bias is a type of
selection bias aka volunteer bias
184
compliance bias applies to people who choose to
participate being healthier or at higher risk of disease than those who don't participate
185
lead-time is the amount of time by which
the dx of a disease was advanced due to screening
186
lead-time bias means
survivals are increased among screen-detested ases only bc dx was made earlier in course of disease
187
length-based sampling bias is a type of
incidence-prevalence bias where less aggressive forms of a disease are likely to be detected b/c of a screening program
188
over-diagnosis bias
Disease has limited malignant/clinical presentation potential
Extreme form of length-based sampling
E.g. PSA testing and low-grade prostate cancer
189
completing risks bias
Missed cases detection due to death unrelated to dx of interest
E.g. prostate CA detection missed in someone who died of CVD
190
Serendipity bias
Chance detection due to dx testing for another reason
E.g. chest x-ray for TB screening finds lung cancer
191
Surveillance bias
aka detection bias
disease ascertainment in monitored pop is better than in gen’l pop
192
recall bias is the
differential recall of exposure b/w cases and control
193
Reporting bias is aka
wish bias, socially desirable responding
194
examples of a real cause
Cigarette smoking → lung cancer
HPV infection → cervical cancer
Asbestos exposure → mesothelioma
194
key characteristics of a real cause
There is a true increase or decrease in disease risk
The association persists after controlling for bias and confounding
It is biologically and epidemiologically plausible
195
what is a real cause?
true etiologic relationship between an exposure and a health outcome
196
what is an artifactual cause?
an apparent association that is not real, produced by problems in study design, data collection, or analysis
197
example of an artifactual cause
Hospital-based case–control study where controls are more likely to smoke than the general population → underestimates smoking–disease association
198
what makes a cause necessary?
if the disease cannot occur without it
199
what makes a cause sufficient?
if its presence always leads to disease
200
Randomization of subjects to treatment groups A and B guarantees that both groups have the same inherent risk of the disease independent of the treatment. T or F?
false
201
high specificity =
high PPV
202
validity compares to
true value
203
reliability is about
consistency
204
higher specificity is more/less false positives?
less false positives
205
lower specificity is more or less false positives?
more false positives
206
lower PPV =
more false positives
207
higher PPV =
less false positives
208
sensitivity =
TP / TP + FN
TP = true positive
FN = false negative
208
NPV is driven by
Specificity
Disease prevalence
209
specificity =
TN / TN + FP
TN = true negative
FP = false positive
210
NPV =
TN / TN + FN
TN = true negative
FN = false negative
211
"Out of those who test positive, X truly have disease" -->
PPV
212
“Out of those who truly have disease, X test positive” -->
Sensitivity
213
random sampling is a (x) method
RECRUITMENT
214
randomization is (x) method
STUDY DESIGN
215
validity for random sampling
external validity
216
validity for randomization
internal validity
217
cluster RT are better when...
you are trying to target specific populations
218
selection bias asks
who gets to participate in the study
219
types of bias within selection bias
volunteer bias
survivor bias
healthy participants bias
220
information bias asks
what info is being collected and how accurate
221
information bias can be (x) and (y)
differential and non-differential
222
biases within information bias
recall bias
socially desirable bias
wish bias
surveillance bias
surrogate bias
lead time
223
what is the causal pathway between the exposure and outcome?
mediators
224
the exposure leads to a (x) and the (x) leads to the (y)
mediator
outcome
225
crude OR = 3.00
adjusted OR = 1.50
partial confounding
226
crude OR = 3.00
adjusted OR = 1.00
TOTAL confounding
227
crude OR = 1.50
adjusted OR = 3.00
PARTIAL confounding
228
crude OR = 1.00
adjusted OR = 3.00
TOTAL CONFOUNDING
229
how do we evaluate causal interference
bradford-hill criteria
rothman's causal pies
230
a sufficient cause is a set of
minimal conditions or events that inevitably leads to disease
231
a component cause is any one of a set of...
conditions which are necessary for the completion of sufficient cause
232
primary outbreak
the disease is introduced to the population from the source
233
secondary outbreak:
people infected in the primary outbreak infect others
234
tertiary outbreak:
the people infected in the secondary outbreak infect a third group of people
235
what type of bias is length-time bias?
selection bias
236
sensitivity and specificity are about the
accuracy of the test
237
PPV and NPV are about the
performance of the test
fall 2025
flashcards
Decks in class (3)
# Cards
