Select the first letter of the word from the list above to
jump to appropriate section of the glossary.
- Absolute risk
and its reduction
- This is the percentage of subjects in any group or
sub-group that experiences a discrete bad outcome such as
death or admission to the hospital. An efficacious
therapy serves to reduce that risk. For example, if 15%
of the placebo group died and 10% of the treatment group
died, the absolute reduction in the risk of death is 5%.
- Accuracy
- The proportion of all test results (positives and
negatives) which agreed with the gold
standard.
-
- Applicability
(also called external validity, generalizability,
relevance)
- This is the degree to which the results of an
observation, study, or review are likely to hold true in
your practice setting.
- For a detailed discussion, click HERE.
- Bayes' Theorem
- This is a simple formula that says that if a particular
test result is twice as likely to occur
in patients with a disease, condition, or injury than in
patients without, then, it is twice as likely that the
patient with the result being tested for actually has the
disease as compared to any randomly selected similar
patient who has not been tested. If you don't like
thinking about things like this, just use the nomogram in
the users guides or the calculator on the diagnosis
appraisal page.
- Bias
- This is any factor which might change the results of a
study from what they would have been if that factor were
NOT present. The direction of bias may be unpredictable.
For example, giving a team a ten point advantage might
seem to give that side an advantage but some teams
actually play much better when they have to come from
behind! The validity of a study
is integrally related to the likelihood that the results
have been biased by factors extraneous to the study
design.
- Blinding
- The "masking" or concealment from study
subjects, caregivers, or others involved in the study of
any detail(s) of the study which could introduce Bias. For example, not
telling patients or doctors which patient gets placebo or
actual drug; or not telling radiologists the clinical
assessment of patients whose films they are reading.
-
-
- Case-control study
- This might be considered a randomized controlled trial
played backwards. People who get sick or have a bad
outcome are identified and "matched" with
people who did better. Then, the effects of the therapy
or harmful exposure which might have been administered at
the start of the trial are evaluated. In other words, you
first find the people who did poorly and then look at the
therapy or exposure and compare it to people who didn't
get the therapy. Needless to say, this is a crude way of
doing a study. When the effect of interest is HARM, this
may actually be the only ethical way of doing the study.
-
- Case report
- This includes single case reports and published case
series'. These are searchable as a separate category in
the MEDLINE database
-
- C.A.T.
- see critically
appraised topic
-
- Clinical
significance
- Results are clinically significant when they make enough
difference to you and your patient to justify changing
your way of doing things. For example, a drug which is
found in a megatrial of 50,000 adults with acute asthma
to increase FEV1 by only 0.5% (P value<.0001)
has failed this test of significance.
-
- Cochrane
Collaboration
- An international organized effort to organize all
existing clinical studies into systematic reviews easily
accessible to practicing clinicians and to otherwise
facilitate the process of bringing clinical evidence to
bear on decision making in patient care.
-
- Cohort study
- Also called a "prospective observational
study", this design follows a group of patients,
called a "cohort", over time to determine
general outcome as well as the outcomes of different
subgroups.
-
- Cointervention
- A therapy or other ancillary treatment which is NOT under
investigation which is given to study patients.
-
- Confidence
intervals
- An interval around an observed parameter such as relative risk
which is guaranteed to include the true value to some
level of confidence (usually 95%). That level of
confidence is only justified to the extent that bias is absent from the study. A well
known election poll advertises itself "this poll is
accurate to within 2 percentage points 99% of the
time." This is a way of saying, in language aimed at
voters (perhaps a skewed sample from the standpoint of
IQ) that the 99% CI around the reported percentages is +
2.
-
- Controlled
clinical trial
- Any study which compares two groups by virtue of
different therapies or exposures fulfills this
definition.
-
- Critical appraisal
- The process of assessing and interpreting evidence
systematically considering its validity, results, and
relevance. For more information, consult the User's Guides.
-
- Critically
appraised topic (C.A.T.)
- A 1 or 2 page summary of a search and critical appraisal
of the literature related to a focused clinical question.
This summary should be kept in an easily accessible place
so that it can be used to help make clinical decisions.
- Dichotomous outcome
- Any outcome measure in which there are only two
possibilities, like dead/alive, admitted/discharged,
graduated/sent to glue factory. Beware of potentially
fake dichotomous outcome reports such as "improved/
not improved", particularly when derived from
continuous outcome measures. For example, if I define a
10 point or greater increase in peak expiratory flow in a
study of acute asthma as "improved", I may show
what looks like a tremendous benefit when the results
were clinically insignificant. This is lesson 2a in
"How To Lie With Statistics."
-
- Double blind
- A single blind study means that someone (patient or
physician) does not know what is going on. Double blind
means that at least two people (patient and physician)
don't know what's going on. Triple blind might mean that
the paper is written before the results are tabulated.
The whole point is to prevent bias.
-
- Effect size
- The difference in measured outcomes attributed
to a therapeutic intervention. This term is encountered
in meta-analyses when
different studies have measured different things. For
example, results of an asthma study which measured FEV1
could be combined with those of another study which
measured return visits to the ED using a statistically
derived generic effect size. Do you prefer skiing or red
wine?
-
- Effectiveness
- I buy a BMW which test drives miraculously on the
dealer’s special runway. I then find that the roads
in the area where I live have all been closed. This is a
breakdown of effectiveness. See efficacy.
-
- Efficacy
- The BMW I have selected for a test drive blows all four
tires, stalls out and crashes on the dealer’s
special runway. I spend two days in the hospital. This is
a breakdown in efficacy. See effectiveness.
-
- Event rate
- This is a term for absolute risk.
-
- Exposure
- Anything you can be exposed to: a drug, a surgical
procedure, time, sexual harassment, rounds, even a
diagnostic test. Most commonly encountered in therapy,
prognosis or harm studies where the EFFECT of an
"exposure" is the subject of the study.
-
- External validity
- See applicability.
- (empty)
- Generalizability
- See applicability.
-
- Gold standard
- No longer relevant in the realm of high finance from
whence it originated, this term gained new life when it
was decided that it should refer to a reference standard
for evaluation of a diagnostic test. For the purposes of
a study, the "gold standard" test is assumed to
have 100% sensitivity and specificity. This may well
constitute an exaggerated estimate of the reference test.
Choice of the "gold standard" must therefore be
evaluated in appraising a diagnosis study.
- Harm-Benefit Line
- On a graph of outcomes, this line divides results
favoring therapy from results favoring the control.
-
- Heterogeneity
- Also called "homogeneity" but having nothing to
do with sexual preference, this term is used to designate
a statistical test used to determine whether results from
a set of independently performed studies on a particular
question are similar enough to make statistical pooling
valid. Are the apples sufficiently red and the oranges
sufficiently green to be able to add them up and report
the total number of "orpples"? As in other
matters, statistical tests do not guarantee clinical
relevance.
-
- Homogeneity
- See heterogeneity.
- Incidence
- The rate at which an event occurs in a defined
population over time. To be distinguished from prevalence.
-
- Intention-to-treat
- Intentions... that with which the path to hell is lined.
Patients assigned to a particular treatment group by the
study protocol should be retained in that group for the
purpose of analysis of the study results no matter what
happens. Patients redefined or dropped from a study early
on as a result of protocol violations unlikely to create
bias may validly be considered exceptions to this rule.
-
- Internal validity
- See validity.
- (empty)
- (empty)
- Likelihood Ratio
- An operator defined as the percentage of patients
positive by gold standard
for a particular disease, condition or injury who have a
particular test result divided by the percentage of
patients without the problem who have that same test
result. A likelihood ratio of two means that the test
result in question is twice as likely to come a patient
with the problem as it is from a patient without the
problem. The LR may be derived from reported sensitivity
and specificity or from a clear understanding of the
above definition. To see how the LR is used, see Bayes‘ Theorem; to
actually use it, see the nomogram.
To see how the Likelihood Ratio is generated, use the calculator
- Meta-analysis
- A review of a focused clinical question following
rigorous methodological criteria and employing
statistical techniques to combine data from independently
performed studies on that question. To learn more, see
the User’s Guide.
- Nomogram for Likelihood
Ratio
-
- Null hypothesis
- What do you do when you want others to be maximally
impressed with what you do? You DECREASE EXPECTATIONS,
then what you do accomplish looks even better! The null
hypothesis is the assumption that there is no difference
between the groups and that the treatment you are
studying has no effect. Any difference in outcome
actually observed between the groups is then evaluated in
relationship to the "zero expectation"
hypothesis.
-
- Number needed to
treat (NNT)
- The number of patients who must receive a particular
therapy for one to benefit. You might tell a patient that
an NNT of 10 means that the chance that he/she will
benefit in this way from the treatment is 1 in 10. To
calculate NNT use the calculator.
- Observational study
- Any study of therapy, prevention or harm in which the exposure is not
assigned to the individual subject by the
investigator(s). A synonym is
"non-experimental"; examples are case-control
and cohort
studies.
-
- Odds ratio
- The odds of an event, understood best by those who enjoy
wagers, is the number of times it occurred (a)
divided by the number of times it didn’t (b), or
a/b. This contrasts with the probability of an
event which is the number of times it occurred divided by
the number of times it could have occurred, or a/a+b.
The odds ratio is the ratio of the odds of an event in
one group divided by the odds in another group. When the event rate or absolute risk in
the control group is small (less than 20% or so), then
the odds ratio is very close to the relative
risk.
- Placebo
- The thing you give a study subject who has been assigned
to the control group to make them think they are getting
the treatment you are studying.
-
- Point estimate
- The exact result that has been observed in a study. The confidence interval
tells you the range within which the result is likely to
lie.
-
- Post-test probability
- The likelihood that your patient has the disease,
condition or injury you are testing for at the moment the
result of the test you (or someone) ordered is delivered
to you. To calculate it you need the pretest probability or prevalence and also the likelihood ratio for the
test in question. To do this, you could use Bayes theorem or, if you are
lazy (and practical), use the nomogram.
-
- Pre-test probability
- At the point you order a diagnostic test, you already
have some idea of how likely your patient is to have the
disease, condition or injury in question. You think of
this as small, medium or large. "Pretest
probability" means putting a number on the estimate
you have already made. A difference of 10% in either
direction will not change the effect of the diagnostic
test. Putting the number on your clinical estimate will,
however, allow you to determine what the test result
means, should you want to know. This is also called prevalence.
-
- Prevalence
- The proportion of people in a defined group who have a
disease, condition or injury. In the context of
diagnosis, this is also called "pre-test probability."
To be distinguished from incidence.
-
- Prospective study
- Any study done forwards in time. This is particularly
important in studies on therapy, prognosis or harm, where
retrospective studies make hidden biases very likely.
-
- Publication bias
- A possible bias which can effect systematic overviews to
the extent that studies on the question at hand with
conflicting results may not have been published.
-
- P value
- The probability that the difference(s) observed between
two or more groups in a study would occurred if there
were no differences between the groups other than those
created by random selection. The assumption underlying
the p-value is the null
hypothesis.
-
- Power
- The chance that an experimental study will correctly
observe a statistically significant difference between
the study groups. This may be considered the
"sensitivity" of the study trial itself for
detecting a difference when it is there.
- (empty)
- Randomization
- A technique which gives every patient an equal chance of
winding up in any particular arm of a controlled clinical
trial.
-
- Randomized
Controlled Trial
- A controlled
clinical trial in which the study groups are created
through randomization.
-
- Relative
risk and its reduction
- The probability of an event in one group divided by the
probability of the same event in another group. Generally
the event is a bad one and the rate in the therapy group
(when it is a therapy study) is in the numerator. When a
benefit has been observed, this ratio is less than one.
Subtracting the ratio from one gives the relative risk
reduction, which is the percentage by which the risk in
the control group has been reduced by the therapy.
-
- Reliability
- Sometimes used loosely, this actually refers to the
reproducibility of a measurement procedure. It is NOT
the same as validity or applicability of a
study.
-
- Retrospective study
- Any study in which the outcomes have already occurred
before the study has begun.
-
- Risk factor
- Any aspect of an individual’s life, behavior or
inheritance which increases the likelihood of a disease,
condition or injury.
- Sensitivity
- The probability that a patient with a disease, condition
or injury will test positive by a particular test for the
problem.
-
- Sensitivity analysis
- An analytical procedure to determine how the results of a
study would change if the facts were different or
different studies included. This is chiefly important in meta-analysis or complex
techniques such as decision analysis and
cost-effectiveness analysis.
-
- Specificity
- The probability that patients without a particular
disease, condition or injury will test negative for the
problem by a particular test.
-
- Statistical power
- see Power
-
- Statistical
significance
- A measure of how confidently an observed difference
between two or more groups can be attributed to the study
interventions. The p value is the
most commonly encountered way of reporting statistical
significance. The methods assume that the study is free
of bias. Clinical
significance is entirely independent from statistical
significance.
-
- Stratified
randomization
- A way of ensuring that the different groups in an
experimental trial are balanced with respect to important
factors which could effect outcome.
-
- Spectrum
- In a diagnosis study, the range of clinical presentations
and of relevant disease advancement exhibited by the
subjects included in the study.
-
- Systematic overview
- A formal review of a focused clinical question based on a
comprehensive search strategy and structured critical
appraisal.
-
- Threshold Probabilities
- The level of suspicion at which your clinical decision
changes
- User's Guide
- A guide to using literature that was developed by the
evidence based medicine group from McMaster University in
Canada. For further information, please click on McMaster
link on main menu.
-
- Utility
- Particularly for a diagnostic test, this is a measure of
whether the patient is truly better off as a result of
the test. A test could have high sensitivity,
specificity and good likelihood ratios and still
have low utility if it is very invasive or poses other
risks or inconvenience to the patient. It belongs under
the section of a diagnostic review.
- Validity
- The degree to which the results of a study are likely to
be true, believable and free of bias. This is entirely
independent of the precision of the results (p value) and does not predict the of
the results to your patients. For a detailed discussion
of validity, click HERE.
- (empty)
- (empty)
- (empty)
- (empty)