Bias The introduction of a  Selection bias

systematic difference o Systematic differences in the characteristics of

between groups being the groups being studied
studied OR between those o Systematic differences between selection and
selected for the study and control groups = decreased internal validity
o Systematic differences between those selected
those who were not
for the study and those who were not =
decreased external validity
 Measurement bias
o Systematic differences between groups in the
way data is collected
o Misclassification is a source of measurement
bias ONLY if the misclassification unevenly
affects the groups
Covariate Any variable that is associated with the outcome. These are often distinguished from
confounders only by the context (i.e. when a variable is placed in a multiple regression
model it is called a covariate, but when a variable is purposefully distributed evenly
between treatment and control so as to eliminate its effect on the outcome, it is a
confounder)
Confounder A variable that is
associated with the Confounder
exposure and also
influences the outcome
(but independently of the Exposure Outcome
causal connection …causes…
between exposure and
outcome)
(Point) Prevalence Number of existing cases cases
¿ ×100 %
population
Incidence Number of new cases in a
cases
population, in a given time ¿ ×100 % per time interval
population
interval
Incidence Rate Ratio Incidence in population 1 : cases1 cases2
incidence in population 2 ¿ ( population1
× 100 % :)(
population 2 )
×100 % per time
interval
Absolute/attributabl
What is the difference in
e risk risk between treatment ¿ incidence∈treatment−incidence ∈control
and control?
Relative risk How many times less or
more likely is it that the
primary outcome will
incidence∈treatment
occur to a person in the ¿
incidence ∈control
treatment arm, compared
to someone who is in the
control?
Number Needed to How many people would I
Treat need to treat with the
intervention, in order to 1
¿
prevent one instance of absolute risk
the primary outcome in a
given time interval?
Hazard A hazard rate is not
measured on its own,
because it is an occurences of the outcome
¿ in a given time interval (e.g.
instantaneous incidence number at risk
rate for a small period of a week1, week2)
study – thus, each hazard
rate is just one in a series.
Hazard Ratio The ratio of hazard rate in
treatment versus control,
over the whole period of =how many times more/less probable the outcome is in the
the study (it is not simply treatment versus control (regardless of what point in the study we
the ratio of the hazard look at)
rates at one particular
point)
Number at risk The number of people for
whom the outcome can
still occur (i.e. everyone
who is still in the study
and to whom the outcome
has not yet occurred)
Odds The ratio of exposed to
exposed probability
unexposed in a case ¿ =
unexposed 1−probability
control study
Odds ratio The ratio of odds for those
odds (with x)
with a certain outcome ¿ where x is the condition of interest (e.g.
versus those without (case odds( without x )
breast cancer)
control study)
Type 1 error (α) “False positive” – showing Given by the p-value. Generally, a positive result is accepted only
an affect when there if this is less than 0.05 (i.e. the probability of this positive result
actually isn’t one being a “false positive” is less than 5%)
Type 2 error (β) “False negative” – showing
¿ 1− power (in a study that has a power of 85%, there is a
that there is no affect,
1-0.85=15% chance that a negative result will be a “false negative”
when there actually is one
p-value The probability that the
Usually, we have no evidence to reject the null hypothesis if this is
result obtained is not due
greater than 0.05 (i.e. there is a greater than 5% chance that what
to type 1 error (it is a true we really have is a type 1 error)
positive)
Sample size n=2 ¿¿
calculation δ =minimum clinically important difference
(comparison of two σ =standard deviation
means) Z α ∧Z β depend on desired α ∧β
Causation  Temporality - cause precedes effect. e.g. smoking precedes lung cancer
 Strength of association e.g. risk of lung cancer in smokers is over 20 times that of non-
smokers
 Dose-response relationship e.g. heavier smokers more likely to develop lung cancer
 Reversibility of the relationship e.g. quitting smoking reduces likelihood of developing
lung cancer
 Consistency e.g. several studies at different times, in different settings have shown the
same association
 Biologic plausibility e.g. compounds in cigarettes carcinogenic in vitro
Internal validity The extent to which the Depends on:
results of the study validly  Study design (minimisation of bias via things like
apply to the sample of people randomisation and blinding)
being studied  Analysis (appropriate use of analysis, analysis that takes
into account confounding, meets criteria for statistical
and clinical significance, low number needed to treat etc.)
External validity The extent to which the Depends on:
results of the study validly  Whether the PICOT question answered by the study
apply to the range of patients reflects the PICOT question concerning a patient
in practice  OR, can my PICOT question be answered using a subset
of the study sample (e.g. the average age of the study
was 64 and my patient is 55, but the study also has a
table where the results are separated out for different
age brackets)
PICOT Population
Intervention/exposure
Control
Outcome
Timing
Specificity The percentage of people true negatives
who truly do not have the ¿
disease and test negative true negatives+ false positives
Sensitivity The percentage of people true positives
who have the disease and ¿
test positive true positives+ false negatives
Positive predictive % people with positive test true positives
that truly have disease ¿
value true positives+ false positives
Negative predictive % people with negative test true negatives
that do not have disease ¿
value true negatives+ false negatives
Positive likelihood Likelihood that a positive
ratio result would be present in a sensitivity
patient with the disease, ¿
compared to a patient (1−specificity)
without the disease
Negative likelihood Likelihood that a negative
ratio result would be present in a
(1−sensitivity)
patient with the disease, ¿
compared to a patient specificity
without the disease