Glossary of Terms


Select the first letter of the word from the list above to jump to appropriate section of the glossary.

- A -

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%.
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.
Back to Top

- B -

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.
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.
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.
Back to Top

- C -

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
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.
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.
Back to Top

- D -

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.
Back to Top

- E -

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?
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.
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.
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.
Back to Top

- F -

Back to Top

- G -

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.
Back t o Top

- H -

Harm-Benefit Line
On a graph of outcomes, this line divides results favoring therapy from results favoring the control.
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.
See heterogeneity.
Back to Top

- I -

The rate at which an event occurs in a defined population over time. To be distinguished from prevalence.
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.
Back to Top

- J -

Back to Top

- K -

Back to Top

- L -

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
Back to Top

- M -

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.
Back to Top

- N -

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.
Back to Top

- O -

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.
Back to Top

- P -

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.
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.
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.
Back to Top

- Q -

Back to Top

- R -

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.
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.
Back to Top

- S -

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.
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.
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.
Back to Top

- T -

Threshold Probabilities
The level of suspicion at which your clinical decision changes
Back to Top

- U -

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.
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.
Back to Top

- V -

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.
Back to Top

- W -

Back to Top

- X -

Back to Top

- Y -

Back to Top

- Z -

Back to Top