About
PMID Numbers: We frequently utilize a PMID number in
place of a citation. Where PMID numbers are available, enter that
number into the PubMed search box to retrieve that citation and
listing.
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Absolute
risk reduction (ARR) – A measure of outcomes which reflects
the actual percent difference in study outcomes between groups.
Formula: % in comparison group  % in study group
Example: Mortality in control group 8%; mortality in intervention group 5%; ARR=8%5%=3%
Similar concept: Absolute risk increase (ARI).
Related: Measures of outcomes
Accuracy –
The ability to correctly identify that which one intends to identify.
Active agent —
Active drug (i.e., not the placebo).
Adjustment  Statistical
means for minimizing differences in study and comparison groups
(e.g., standardization and regression analysis).Adverse event
(ADE) — see Safety
Aim, study —
see Objectives
Allocation concealment
— see Concealment of allocation
Allornone results  Very large results between groups or beforeandafter results—for example, observed outcomes in which, before the intervention, nearly everyone dies, but following use of the intervention, nearly everyone survives.
There is general agreement that causality MAY be present when there is dramatic change following application of an intervention or technology that is unlikely to be due to confounding (i.e., close to allornone results: example, before treatment all died and following treatment, high survival rate). Examples of allornone results are antibiotics for meningitis and electrical cardioversion for ventricular fibrillation.
Alpha error —
see Error
Alpha spending  see Multiplicity adjustment or testing
A priori – In advance. For example, issues such as study questions,
outcome measures, subgroups for analysis and pvalues should be
determined in advance of the actual study (i.e., determined a
priori). Reduces risk of chance findings for prespecified outcomes.
Ascertainment bias
 See bias
Assessment bias 
See bias
Association –
Statistical relationship between two or more events, characteristics
or other variables. Example: weight is related to height.
Attrition bias 
See bias
Baseline
characteristics — Key characteristics of interest at the
start of a study which are potential prognostic factors. These
usually include demographic information such as age, gender, race
and other characteristics related to the area under study such
as disease severity, baseline measurements for clinical items
of interest, etc. Collecting and reporting on baseline characteristics
helps to establish how similar the study groups are, describes
the actual population studied and helps to measure improvements
during the course of the study.
Benchmark –
A standard or point of reference used in measuring and/or judging
something such as quality or value.
Beta error  see
Error
Bias – Study
processes or events which may result in, or lead to, conclusions differing
from truth in a systematic way (meaning not due to chance). Bias
may occur at various study stages such as assigning subjects to
study or comparison groups, intervention or exposure, performance,
provision of services or conduct of processes affecting subjects,
data collection, subject followup, measurement, analysis, interpretation
and/or publication of data. Bias frequently occurs as a result
of some inequality between the study and the comparison group.
The inequality may be important because it may affect the study
outcome.
Key Biases Which
May Occur at Various Study Stages
Selection bias
– A bias occurring in subject identification, selection
and/or assignment. This includes dissimilarity between groups.
Observation bias
– Bias that may have been introduced through study procedures
during implementation, intervention or exposure, followup or
assessment. Specific biases in this category primarily consist
of –
 Performance
bias – Threats to validity, postrandomization, arising
from study activities involving subjects such as interventions,
processes or procedures. Frequently this is a bias that can
result when there are differences between groups other than
the intervention under study.
Synonym: Intervention bias
 Attrition bias
– A bias occurring as a result of subjects lost to followup
through withdrawals or other study attrition.
Synonym: Followup bias
 Assessment
bias – A bias occurring in the way outcomes are assessed.
Synonyms: Ascertainment bias, Detection bias, Measurement
bias
Bivariate analysis  Bivariate analysis involves the analysis
of two variables to show relationships between those two variables.
Data from two variables are plotted on a joint distribution graph,
and then correlation analysis and simple linear regression analysis
are performed to assess how closely the variables are related.
Related: Pearson correlation coefficient
Blinding –
An important study procedure to keep secret certain study procedures
such as which is an active drug and which is a placebo. Bias can
result when study subjects and those involved in study procedures
know treatment assignment of individual subjects. Blinding is
a method to help avoid the introduction of this kind of bias.
Doubleblinding refers to when neither patient nor persons performing
the intervention or exposure — or working with the subjects'
data — know if the patient is in the study group or the
comparison group.
Synonym: Masking
Related: Concealment of allocation, Doubleblinding, Doubledummy,
Encapsulation
Case
— Research subject who is a member of the group possessing
the characteristic under study such as an exposure, risk factor,
outcome or receiving the intervention in an experimental study.
Casecontrol study
– An observational study in which subjects with outcomes
of interest (cases) are compared to those without the outcome
(controls). Histories are examined to attempt to determine exposures.
Case series –
A group of patients receives an intervention and outcomes are
assessed. There is no comparison group. Case series is almost
never “evidence.” Allornone results are required,
which rarely happens.
Case study —
An observational of one person. Also referred to as a case report.
A case study is not evidence.
Censoring  Censoring is the practice of removing the patient from the curve at a specific point in time. Examples of censoring: 1) Patients who don’t experience the event (administrative, or right, censoring, which is appropriate), 2) Other reasons determined by the investigators and called “censoring rules” (nonadministrative censoring such as lost to followup or dying before a nonmortality outcome of interest is reached), and which must be critically appraised.
Chance — Research
results happening by "accident," meaning that they did
not happen because of truth or because of bias. Likelihood is
assessed by using appropriate statistical tests. To critically
appraise, examine pvalues and confidence intervals with associated
pointestimates (none of which can address bias, which is a systematic
error  meaning an error not due to chance).
Synonyms: Random error, Sampling error
Related: Bias
Class effect  Class
effect refers to a concept that certain drugs share sufficient
similarities that they can be thought of as a class. It is a determination
that a set of agents with similar chemical structures, mechanisms
of action and pharmacological effects has similar therapeutic
and adverse effects. There are no universally accepted criteria
for defining class effect. (For all intents and purposes, class
effect treats the agents as if they are the same clinically, but
some other factors, such as cost, may vary.)
Clinical practice
guidelines – Clinical practice guidelines are systematically
developed statements to assist practitioners and patients in choosing
appropriate healthcare for specific conditions. They should be
based on valid and useful evidence, and they should be critically
appraised before adoption.
Clinical significance
— Research should benefit patients in ways that are clinically
significant (i.e., that matter to them). Clinically significant
areas are morbidity, mortality, symptom relief, functioning and
healthrelated quality of life. Results from valid research in
these areas should be large enough and useful enough to provide
meaningful clinical benefit.
Clinical trial —
A research methodology which evaluates a drug or other intervention
to assess safety and efficacy by evaluating its effect on a group
of subjects. The subjects are divided into two or more groups
with the study group receiving the drug or intervention and the
comparison group or groups receiving a comparison drug, placebo
or intervention.
Cohort study –
An observational study in which exposure occurs in the study group
and not in the comparison group. Exposure is not introduced by
the investigator, but happens through other means such as natural
occurrence or individual choice. Groups are followed over time
and outcomes are compared. Cohort studies can be done prospectively
where a group of interest and a comparison group are followed
forward in time. Cohort studies can be done retrospectively when
the group of interest and comparison group are identified from
the past, based on an exposure or characteristic, and then information
about them is reviewed from that time forward to assess development
of outcomes.
Comparison groups
— Subjects in groups being compared to the group of interest
(e.g., intervention group would be “group of interest”
and placebo group would be the “comparison group”).
Composite endpoints
— Refers to an endpoint in which single endpoints are grouped
together to form one endpoint such as “cardiovascular events.”
Concealment of allocation
– The process used in a randomized controlled trial to hide
the assignment to a study group. The purpose is to ensure that
no one can influence or control which study subject gets assigned
to which group (e.g., assignment made through a callcenter, etc.)
Related: Blinding
Confidence interval (CI) – CIs represent a range of statistically plausible results consistent with an outcome from a single study. Confidence intervals have some practical limitations similar to Pvalues (see Pvalues). Although the CIs can project a range of results consistent with the study results, they cannot tell you the truth of the outcomes. CIs cannot replace the need to critically appraise the study.
Our approach:
Despite their limitations, we believe confidence intervals to be more informative than Pvalues. We approach them as providing a possible range of plausible results for the larger population IF the study results in the studied population are true.
For valid studies, we make a judgment for what we consider to be a reasonable range for clinical significance—this need not be hard and fast. For statistically significant findings, is the confidence interval wholly within bounds for clinical significance? For nonsignificant findings, is the confidence interval wholly beneath your limit for clinical significance? A yes to these two questions means likely conclusive findings for valid studies. No, means findings are inconclusive.
Reporting styles vary. Our favorite is below.
Example: ARR 5%, 95% CI (3% to 7%)
(Avoid using "" when separating the interval numbers, which can be confused with a minus sign.)
Related: Pvalue
Confounding –
A special type of bias in which another factor associated with
the study variable of interest may have "traveled" with
that variable and is the true reason for the study conclusion
instead of the variable under study or also affects the study results. Example: HRT users were found to have lower second MI rates; however, the results were confounded by HRT users living healthier lives. There are known confounders
and unknown confounders. Randomization is a method which attempts
to minimize confounding by randomly allocating subjects to their
groups in hopes that any potential confounders are equally distributed
between the groups.
Contamination –
Any member of a comparison group receiving the intervention under
study or being exposed to the variable of interest.
Control  Research
subject who is a member of the comparison group.
Control Event Rate (CER)  The proportion of patients in a comparison (control) group in whom an event is observed.
Controlled clinical
trial – An experiment in which investigators include a control
group which is used as a comparison group. The comparison group
may receive another treatment, a placebo or “usual care”.
Subjects are assigned to the study or to the control group. Intervention
or exposure is then performed. Patients are followed to determine
outcomes such as improvements or harms.
Related: Clinical trial, Randomized controlled trial (RCT)
Correlation  Variables that are related to each other such that, when observing a correlation between the two variables in a graphic display, when one variable changes the other variable is seen to change. The changes cannot be assumed to be causal. Example:in humans, weight is positively correlated to height.
Covariate  A variable
that might affect, or possibly predict, an outcome.
Critical appraisal
– A scientific evaluation of evidence (e.g., research data)
to appraise validity (closeness to truth) and usefulness (e.g.,
generalizability to one's own patients or circumstances, meaningful
benefit, etc).
Crossover design
 A study design in which patients start out having one intervention
and then are “crossed over” to a different intervention.
Crossover trial –
A comparison of two or more interventions in which the participants
receive one treatment and then are “crossed over”
to another intervention. Care must be taken with this study design
because effects of the first treatment could contaminate the effects
of the subsequent intervention.
Related: Contamination
Crosssectional study
– An observational study in which a sample from a larger
population is examined, at one single point in time – like
a snapshot  for prevalence of an exposure or characteristic along
with an outcome of interest. Crosssectional studies are also
frequently used to evaluate diagnostic testing.
Database
research — Statistical relationships between two or more
variables are assessed from databases. Frequently database research
is used to report outcome measures. Database research, however,
is a type of observation and, like all observational research,
is highly prone to selection bias, observation bias and confounding
— and the likelihood of finding "statistically significant"
relationships merely due to chance is high to certain, depending
upon the way the analysis is conducted. Definite cause and effect
conclusions should not be drawn from database research.
Data gathering validity
– Data gathering validity refers to the methods used to
obtain numerators and denominators during the data gathering phase
of performance measurement.
Delta — See
Equivalence trials and Noninferiority trials
Denominator –
The lower portion of a fraction used to calculate a rate or ratio.
In a rate, the denominator is usually the population (or population
experience, as in personyears, etc.) at risk.
Related: Numerator
Denominator –
For performance measurement, the population
“at risk” for experiencing the event or occurrence
described in the numerator — the "pool."
Dependent variable
– In a statistical analysis, the outcome variables under
study.
Related: Independent variable, variables
Detection bias 
See bias
Diagnostic testing
— see Measures of test function
Disease spectrum
— A range of symptoms, signs, lab results and results of
other diagnostic tests, and rate of disease progression, response
to therapy and disease severity.
Doubleblind –
see Blinding
Doubledummy treatment
design — A treatment design involving two placebos. Control
subjects receive one or the other of two placebos, but not both.
Study subjects receive the intervention drug. For example, if
you were studying an antibiotic being given at two different doses,
you might need two placebos because of the weight and size difference
of the doses.
Drug utilization
review (DUR): A system for evaluating appropriateness of drug
therapy. Prescribing patterns are evaluated to determine if drugs
are being misprescribed, possibly resulting in problems with safety
or effectiveness.
Effect size  The size of the difference in outcomes between groups.
Effectiveness
– The extent to which a given intervention is likely to
produce beneficial results for which it is intended in ordinary
circumstances.
Related: efficacy
Efficacy –
The extent to which a given intervention is likely to produce
beneficial effects in the context of the research study.
Related: Effectiveness
Empirical –
Empirical results are those which are based on experience or observation.
Encapsulation —
A drug blinding method whereby drugs (active or inert) are crushed
and put into gelatin capsules to disguise the active agent as
compared to the placebo.
Related: Blinding
Endpoint —
see Outcome measure
Epidemiology –
A study of the distribution and determinants of healthrelated
states or events in specified populations, and the application
of this study to the control of health problems.
Equipoise In healthcare,
a state in which it is unknown which interventions will result
in best outcomes.
Equipotency  See
Class effect
Equivalence trial
 Equivalence trials are usually used to demonstrate that the
effects of two treatments do not vary more than a prespecified
clinically acceptable amount and can therefore be considered clinically
equivalent. Delta is the name given to the range of results within
which results are judged to be equivalent.
Error —
 Type 1  or alpha
error  A difference is reported, but there is no difference.
This can be due to bias, confounding or chance.
Related: Statistical significance
 Type 2  or beta
error  No difference is reported, but there is a difference.
This can be due to an insufficient number of people studied.
Related: Power
Estimate of effect
– see Measures of outcomes
Event rate – see Control Event Rate
Evidencebased medicine
(EBM) – Delfini definition: "The use of the scientific
method and application of valid and useful science to inform health
care provision, practice, evaluation and decisions."
Experimental study
– A study in which the investigator controls an exposure
or intervention, then follows subjects to compare outcomes between
research subjects who are exposed or who receive the intervention
and those who are not exposed or who do not receive the intervention.
Tip: If an intervention is "assigned" through the research,
it is an experiment. If it is chosen, then the study type is observational.
Related: Observational study
Exposed group –
A group whose members have been exposed to a supposed cause of
disease or health state of interest, or possess a characteristic
that is a determinant of the health outcome of interest.
External validity
– Whether a study's results are generalizable either to
a patient population (see Population bias) or to "real world"
circumstances (see Intensity bias). Also referred to as
Synonym: Generalizability
Related: Validity, Internal validity, Population bias, Intensity
bias
Firstline
therapy: The drugs to be utilized first in treating a patient.
Followup bias 
See bias
Foraging tools 
Tools that assist with keeping uptodate with new information.
Related:Hunting tools
Forest Plot  A graphic representation of the results of individual trials in a metaanalysis along with a summary diamond. Horizontal lines represent trials; a vertical line represents the line of no difference. The graphic display is useful in evaluating and summing numerous trials at one time.
Formulary: List of
therapeutic agents available in a particular practice for
caring for patients. The term “preferred drug list”
is also used.
 An open formulary
may have few or no restrictions.
 A closed formulary
has restrictions.
Formulary system:
A system that provides for the processes for establishing and
managing the formulary.
Generalizability
– see External validity
Generic substitution:
Replacement of one agent with a different agent having the same
chemical structure. This may be done when the patent on a brandname
drug expires. Bioequivalence is frequently assumed (i.e., it is
assumed that the generic agent is equivalent to the brandname
drug). In some cases the effects of other components of the generic
preparation (e.g., the vehicle in a dermatological preparation)
may vary and result in outcomes that differ from those reported
for the brandname agent.
Gold standard –
The intervention generally believed to be the best available against
which new interventions should be compared.
Halo
effect  How an observer's perception, knowledge or recollection,
may bias results. Also used as a synonym for Placebo effect when
medical attention and services can bias results.
Hawthorne effect
 How the knowledge of being in a study can influence a subject's
behavior (usually favorably) which can bias results.
Hazard  An incidence rate in a survival curve.
See Survival Analysis
Hazard rate  A measure of how rapidly subjects are experiencing the endpoint in a survival analysis
Hazard ratio  (Calculated using Cox proportional hazards model) approximates the relative risk in the intervention group compared to the control group in a KaplanMeier model and is assumed to remain constant (often not a valid assumption).
Heterogeneity  In systematic reviews, incompatibility between trials included in the review.
Hunting tools 
Tools which aid the process for searching to answer a clinical
question.
Related: Foraging tools
Hypothesis —
Tentative explanation that forms the basis of a research study.
Related: Null hypothesis
Imputation
— As in "imputation of missing variables." Principles
of IntentiontoTreat (ITT) analysis require analyzing all patients
in the groups to which they were assigned. Investigators are to
"impute" or assign outcomes for missing data points.
Example: worst case scenario in which study subjects with missing
outcomes are assigned as "treatment failures," and comparison
subjects, assigned as "successes." Frequently "lastobservationcarriedforward"
(LOCF) is used, however, this has been shown to be a method prone
to bias. Imputing outcomes for missing data points is not a method
for "determining the truth of what may have happened if we
had no missing data," but rather a method to test the strength
of the outcomes (i.e., statistically similar results to those
reported) considering all the data points in such a way that does
not favor the intervention. LOCF would be especially biased in
patients with a progressive illness, for example, because patient
outcomes would appear better than would be expected in a progressive
illness. However, there are instances in which LOCF can help support claims of efficacy. See— http://www.delfini.org/delfiniClick_PrimaryStudies.htm#LOCFhelp
Related: Intentiontotreat
Incidence –
The proportion of new cases of the target disorder in the population
at risk during a specified time interval.
Related: Incidence rate, Prevalence
Incidence rate (IR) –
A measure of the frequency with which an event, such as a new
case of illness, occurs in a population over a period of time.
The denominator is the population at risk; the numerator is the
number of new cases occurring during a given time period. Incidence rate is calculated by taking the number of events divided by the persontime at risk.
Incidence rate ratio (IRR)  Incidence rate ratio is the ratio of two incidence rates. The incidence rate is defined as number of events divided by the persontime at risk. To calculate the IRR, the incidence rate among the exposed proportion of the population, divided by the incidence rate in the unexposed portion of the population, gives a relative measure (IRR) of the effect of a given exposure and approximates the relative risk or the odds ratio if the occurrences are rare.
Inconsistency statistic
 The I2 statistic is a test of heterogeneity. The range of I2 values is between 0% and 100%. I2 provides an estimate of the percentage of variability in results across studies that is likely due to true differences in treatment effect as opposed to chance. When the I2 is 0%, chance provides a satisfactory explanation for the variability in the individual study point estimates, and clinicians can be comfortable with a single pooled estimate of treatment effect in a valid study. As the I2 increases, bias becomes more likely. A rule of thumb characterizes an I2 of less than 25% as small heterogeneity, 25% to 50% as moderate and more than 50% as large heterogeneity.
Independent variable
– An exposure, risk factor, or other characteristic being
observed or measured that is hypothesized to influence an event
or outcome (i.e., the dependent variable).
Related: Dependent variable, variables
Inference, statistical
– In statistics, the development of generalizations from
sample data, usually with calculated degrees of uncertainty.
Intensity bias 
An external validity bias in which the circumstances of the study
differ meaningfully from the circumstances to which study methods
might otherwise be applied (e.g., "real world" settings
as contrasted with the highly controlled environment often found
in controlled trials).
Related: External validity
Intentiontotreat
(ITT) – Analyzing results for all patients in the groups to which
they were assigned whether or not they received or completed the
intervention or experienced the exposure. The number randomized
to each group should equal the number analyzed in each group —
and they should be the same people.
Related: Imputation
Intermediate outcome
markers – Outcome measures, such as a biologic factor (biomarker)
or lab/imaging test, that are “assumed” to represent
clinical outcomes (e.g., blood pressure used as a surrogate end
point in studies of stroke).
Synonyms: Proxy markers, Surrogate markers, Surrogate end points
Internal validity
 Closeness to truth within the context of the study (i.e., truth
of the study not taking into account external validity). Assessing
internal validity entails "ruling out" bias, confounding
and chance as possible explanations for an observed association
between an element of interest in a study and resulting outcomes.
Related: Validity, External validity, Bias, Confounding, Chance
Interventions —
Includes screening, prevention and treatments. Only valid randomized
controlled trials with useful results should be used to inform
decisions about interventions in these areas.
Intervention bias
 See bias
KaplanMeier methodology—the most commonly used survival analysis in healthcare.
Synonyms: KaplanMeier estimate, KaplanMeier model
Lead
time bias  A bias resulting from a disease found through screening
as compared to when it might otherwise have been detected. This
kind of bias can result in a treatment seeming to be very effective
if the lead time is long (e.g., “increased” survival
time).
Related: Bias
Length time bias
 A bias that can occur when certain characteristics or conditions
under study differ in the speed of progression. This kind of bias
can result in findings favoring screening. An example is tumors.
Fastergrowing tumors causing symptoms will be more likely to
be found outside of screening. Slowergrowing, asymptomatic tumors
will have a longer duration and be less likely to be found outside
of screening. Thus, screening will identify more slowergrowing
asymptomatic tumors which could then result in a conclusion that
screening helps prevent mortality. This kind of bias is most likely
to occur in screening studies and casecontrol studies where prevalence
cases are included, rather than incidence cases because incidence
cases are assumed to be new starts.
Related: Bias
Likelihood ratios
(LR) – Likelihood ratios can be helpful for comparing one
test to another, and results can help rule in or rule out a condition.
Related: Measures of test function, Positive likelihood ratio
and Negative likelihood ratio.
Line of no difference
– The point at which there is no greater benefit or risk
one way or another (meaning the meeting point for "favors
intervention" versus "favors placebo.")
Related Terms: Line of no effect, Infinity (NNT etc), Unity (ratios)
 Point estimates
expressed as percentages — the meeting line is at 0.
Relative risk reduction = 0 equals no difference
Absolute risk reduction = 0 equals no difference
 Point estimates
expressed as ratios — the meeting line is at 1.
Odds Ratio = 1 equals no difference.
Relative risk (also known as Risk ratio) = 1 equals no difference.
Masking
 see Blinding
Matching –
A mechanism to reduce selection bias by attempting to make a study
group and a comparison group as comparable as possible by matching
similar variables such as age and sex, etc.
Mean, arithmetic
– The average.
Measure of association
– see Measures of outcomes
Measurement bias
 See bias
Measurement of risk
– see Measures of outcomes
Measures of outcomes
— Statistics that show the size of differences between the
results from the study groups.
Synonyms and related terms: Measurement of association, Measurement of risk, Estimates
of effect, Point estimates, Effect size, Treatment effect
 Absolute risk
reduction
 Number needed
to treat
 Odds ratio
 Relative risk
 Relative risk
reduction
Measures of test
function — Measures to help determine the accuracy and usefulness
of diagnostic tests.
Synonyms: Indices of accuracy
See individual definitions for the following:
 Sensitivity
 Specificity
 Positive predictive
value
 Negative predictive
value
 Positive likelihood
ratio
 Negative likelihood
ratio
 Posttest odds
 Posttest probabilities
 Numberneededtodiagnose
Median – The
data midpoint.
MeSH  Medical Subject Headings: a list of synonyms or thesaurus of terms used by search databases to index and classify medical information.
Metaanalysis –
A quantitative technique for summarizing results of more than
one study using predetermined criteria. The goal is to provide
a summary estimate of effect based on the scientific weight of
the studies. Metaanalysis can be achieved through use of the
results of individual studies or actually pooling data from those
studies.
Related: Systematic Review.
Mode – The
most frequently occurring value in a dataset.
Monograph  A written
review and analysis, often of a single agent, containing a list
of usage recommendations.
Multiplicity adjustment or multiplicity testing  Methods to attempt to reduce the higher risk of chance effects resulting from multiple analyses and multiple outcomes being tested by apportioning the pvalue. Portions of the pvalue (e.g. if the pvalue is 0.05 and there are 10 assessments, authors may require a pvalue of 0.005 for statistical significance for each assessment) making it harder to achieve statistical significance. In other words, part of the pvalue is "borrowed" for each assessment. This would amount to lower pvalues for the multiple assessments.
Rothman 99 makes some good points about this:
 If the risk for rejecting a truly null hypothesis is, for example, 5% for every hypothesis examined then examining multiple hypotheses will generate a large number of errors simply because of the increasing number of hypotheses examined.
 Adjusting for multiple comparisons is thought by many to result in a smaller probability of erroneously rejecting the null hypothesis, but is that so? Instead, adjusting for multiple comparisons might amount to paying a penalty (decreased power) for simply appropriately doing more comparisons and Rothman argues that we have no empirical justification for adjusting—we do not know whether a given association is an “unpredictable manifestation of random processes” or that we have evidence suggesting that we should “pay for peeking” at more data by adjusting Pvalues.
 Rothman argues that the burden is on those who advocate for multiple comparison adjustments to show we have a problem requiring a statistical adjustment fix.
 Rothman’s conclusion: It is reasonable to consider each association on its own for the information it conveys. [Rothman KJ PMID: 2081237]
Related term: Alphaspending
Multivariable analysis
 A statistical method for determining the specific contributions
of various factors to a single outcome. This allows investigation
of a single variable while controlling for the effect of other
variables. Multivariable analysis determines the independent contribution
of each independent variable to the dependent variable (i.e.,
development of coronary heart disease).
Methods include
multiple linear regression, multiple logistic regression, and
proportional hazards (Cox) regression
 Linear regression
is used with continuous outcomes
 Logistic regression
is used with dichotomous outcomes
 Proportional
hazards regression is used when the outcome is the length of
time to reach a discrete event (such as time from baseline visit
to death)
Multivariate analysis
 A statistical method for determining the specific contributions
of various factors to a more than one outcome.
Narrative
review: An article in the medical literature summarizing other
studies for a given topic, characterized by a lack of a transparent,
scientific, and systematic approach; thus, the summary is highly
likely to be misleading. Instead, systematic reviews should be
sought (and appraised).
Negative likelihood
ratio – The number of times the percent of false negatives
occurs over percent of true negatives.
Related: Measures of test function
Formula = (1sensitivity) / specificity
This represents the change from pretest odds to posttest odds.
The size of an increase is considered as follows —
 Small =.02.05
 Modest = .05.1
 Large = > .1.
Large is considered to rule out a condition.
Negative predictive
value (NPV) – Of all testing negative, percent who do not
have a condition, based on that population's prevalence.
Formula = d / (c + d) from twobytwo table
Related: Measures of test function, Twobytwo table
Network metaanalysis  A method of assessing comparative effectiveness of interventions that have not been directly compared in clinical trials. Network metaanalyses compare the results from two or more trials that have one treatment in common. For example, if treatment A has been compared to treatment B in one trial and in another trial of similar subjects, B has been compared to treatment C, then a network metaanalysis can be used to indirectly compare treatment A to treatment C.
Noninferiority trial
 Noninferiority trials aim to show that an intervention is not
inferior to a comparison intervention by more than a prespecified
clinically acceptable amount (Delta). Judgment is required to
establish what is meant by a clinically acceptable amount.
Nonparametric 
Nonparametric refers to instance in which the distribution is
unknown. Parametric refers to instances in which parameters of
the distribution of the variable of interest is known (e.g., knowledge
that a distribution of a variable is expected to follow a bell
curve).
Null hypothesis –
The first step in testing for statistical significance in which
it is assumed that the exposure is not related to a disease. This
is done for mathematical purposes to level the playing field to
begin with the assumption that study outcomes are the same as
what would have occurred by chance.
Related: Hypothesis
Numberneeded to
diagnose (NND) — From Bandolier: "For any chosen clinical
endpoint the NNT is the reciprocal of the fractional improvement
in a treated group minus the fractional improvement in an untreated
group NNT = 1/(fraction improved with active  fraction improved
with control)" or NND = 1/[Sensitivity  (1  Specificity)].
Primarily useful for comparing the NND values between different
tests. Best outcome is as close to 1.0 as possible. (http://www.jr2.ox.ac.uk/bandolier/band27/b272.html)
Related: Numberneededtotreat, Measures of outcomes
Numberneededtotreat
(NNT) – The number of patients who need to be treated in
order for one patient to benefit over that patient taking the comparator agent within the study time period.
NNT is the reciprocal of the ARR. For an ARR of 5%, for every 20 people treated, 1 more person will have improved outcomes with Agent A than if they were treated with Agent B within the study time.
Formula = 1/ARR (or how many times does ARR # go into 100). Example:
For an ARR of 5 percent the NNT is 20; meaning, twenty people
would have to be treated for one person to benefit.
Similar concept: Number needed to harm / screen / prevent, etc
– NNH, NNS, NNP
Related: Measures of outcomes
Numerator –
The upper portion of a fraction.
Related: Denominator
Numerator –
For performance measurement, the event or occurrence
being tracked (a subset of the denominator) — the "count."
Objective
— Goal or aim of the study.
Observation bias
 See bias
Observational study
– Epidemiological study in which observations are made,
but investigators do not control the exposure or intervention
and other factors. Changes or differences in one characteristic
are studied in relation to changes or differences in others, without
the intervention of the investigator. Observational studies are
highly prone to selection bias, observation bias and confounding.
Tip: If an intervention is "assigned" through the research,
it is an experiment. If it is chosen, then the study type is observational.
Related: Experimental study
Odds  The likelihood
of an event occurring compared to not occurringe.g., odds of
two to one mean that likelihood of an event occurring is twice
that of not occurring.
Odds ratio (OR) –
A point estimate used for casecontrol studies which attempts
to quantify a mathematical relationship between an exposure and
a health outcome. Odds are used in casecontrol studies because
the investigator arbitrarily controls the population; therefore,
probability cannot be determined because the disease rates in
the study population cannot be known. The odds that a case is
exposed to a certain variable are divided by the odds that a control
is exposed to that same variable. Odds are often used in other
types of studies as well, such as metaanalysis, because of various
properties of odds which make them easy to use mathematically.
Odds and probabilities of outcomes are usually similar if the event rate is low, but we recommend calculating both the odds ratio reduction and the relative risk reduction in order to compare and determine if the difference is clinically meaningful.
Related: Measures of outcomes
Open label study
 Not blinded as to the intervention.
Related: Blinding
Outcome — see
Outcome measure
Outcome measure –
What we are interested in studying (e.g., mortality, use of rescue
medications in asthma patients, incidence of no flares in atopic
dermatitis).
Synonyms: Endpoints, Outcomes, Outcome variables
Related: Primary outcomes, secondary outcomes, composite endpoints.
Overdiagnosis bias  A
finding of a disease at an asymptomatic stage in a patient who
would not have become symptomatic or harmed by the disease.
Override  Process
of setting aside a prescriber’s choice of a medication and
usually substituting another medication.
Parametric
 Parametric refers to instances in which parameters of the distribution
of the variable of interest is known (e.g., knowledge that a distribution
of a variable is expected to follow a bell curve. Nonparametric
refers to instance in which the distribution is unknown.
Pathophysiology 
Disruption of normal biochemical or physical function by a disease
or disorder.
Patient years  A way of communicating the rate events in a study. Patientyears is a statistic encountered in clinical studies. It is used because it conveys information about event rates and time. The calculation is accomplished by adding the years that patients are in a study and dividing the years by the number of events. Example: 100 patients are followed for 2 years. In this case, there are 200 patientyears of followup. If there were 8 myocardial infarctions (MI), the rate would be 8 MIs per 200 patient years or 4 MIs per 100 patientyears. Patientyears is used to give information about means. Read more here.
Pearson correlation
coefficient  An analysis method to measure the extent of the
linear relationship of two variables to determine how independent
and dependent variables change together (e.g., salt intake and
blood pressure).
Synonym: r value
Related: Bivariate analysis
Performance bias
 See bias
Performance measure
– A quantitative assessment of a process of care, outcome
or service usually used in quality improvement work. It consists
of a denominator (e.g., population of interest), a numerator (i.e.,
a count of events of interest occurring within the denominator)
and a frequency (i.e., the specified interval for measurement).
May target various levels or units such as a system, specialty
group or individual. Usually expressed as a rate, ratio or percentage.
Sometimes used interchangeably with “quality indicator.”
Key terms = accuracy,
benchmark, data gathering validity, denominator, dependability,
numerator, outcome measure, performance measure, precision,
quality indicator, rate, ratio, risk adjustment, risk stratification
Pharmacy & Therapeutics
Committee: Committee charged with making formulary management
decisions, along with performing other formulary management functions.
Pharmacy benefit
manager (PBM): A company that manages pharmacy benefits and formulary
management for health care systems and/or insurance companies.
Phase (clinical)
studies –
 Phase I is the
first stage in testing a new drug in humans for safety and chemical
action of the drug. Phase I studies are usually performed on
healthy volunteers (often 20 to 100 persons) without a comparison
group.
 Phase II studies
are performed mainly to test a drug’s efficacy. They are
often performed on healthy volunteers (as many as several hundred
people who have the condition in question) and may be up to
two years duration or more. They are sometimes conducted as
randomized controlled trials.
 Phase III studies
are fullscale evaluations of treatment, in hundreds to thousands
of patients, for comparison to the current standard treatments
for the same condition. Phase III studies are frequently randomized
controlled trials.
 Phase IV studies
are postmarketing surveillance studies.
Placebo – An
inactive substance or procedure administered to a patient for
comparison to the intervention. Used in clinical trials to blind
people to their treatment allocation.
Related: Blinding
PICOTS – Acronym standing for opulation, intervention, comparison, outcomes, timing, setting. PICOTS is useful in formulating a clinical question, searching*, comparing and reporting on studies. (*We find searching can be too limited using PICO or PICOTS. We frequently use condition/intervention.)
Pivotal trial  A
controlled trial to evaluate the safety and efficacy of a drug
in patients and usually the basis for the New Drug Application
(NDA) filing with the FDA.
Point estimate –
see Measures of outcomes
Population –
For research studies, the term "population" may refer
to a group of people living in a specific geographic area or it
may mean a unique set of individuals who share some characteristic
such as exposure to a disease or a group of patients of a single
clinician, as examples. In statistics, this can mean a population
of data.
Population bias
 An external validity bias in which the population under study
differs meaningfully from the population to which the results
might be applied.
Related: External validity
Positive likelihood
ratio (LR+) – The number of times the percent of true positives
occurs over percent of false positives. This represents the change
from pretest odds to posttest odds.
Related: Measures of test function
Formula = sensitivity / (1specificity)
This represents the change from pretest odds to posttest odds.
The size of an increase is considered as follows —
 Small = 2 to 5
 Modest = 5 to
10
 Large = > 10.
Large is considered to rule out a condition.
Positive predictive
value (PPV) – Of all testing positive, percent who have
the disease, based on that population's prevalence.
Formula = a / (a + b) from twobytwo table.
Related: Measures of test function, Twobytwo table
Posttest odds for
positive test — The odds of a person having a condition
if the test is positive.
Formula = pretest odds x positive likelihood ratio
Related: Measures of test function
Posttest odds for
negative test — The odds of a person not having a condition
if the test is negative.
Formula = pretest odds x negative likelihood ratio
Related: Measures of test function
Posttest probability
for a positive test — After learning test result, the probability
that a person testing positive has the condition.
Formula = Odds post (T+) / (1 + Odds post (T+))
Related: Measures of test function
Posttest probability
for a negative test — After learning test result, the probability
that a person testing negative has the condition.
Formula = Odds post (T) / (1 + Odds post (T))
Related: Measures of test function
Power  See Statistical
power.
Precision –
The ability to provide sufficient detail, such as small incremental
units, to be useful.
Predictive value
– A diagnostic measure of function which is used to address
the probability of having or not having a condition. Predictive
value is affected by prevalence.
Related: Positive predictive value, Negative predictive value.
Pretest likelihood
— Pretest probability of having a condition based on the
prevalence of the condition in the population studied.
Related: Prevalence
Pretest odds —
Pretest odds of having a condition based on the prevalence of
the condition in the population studied.
Formula = prevalence / (1prevalence)
Related: Prevalence
Pretest Probability The probability that a patient has the condition prior to administering a test.
Prevalence –
The number or proportion of cases or events or conditions in a
given population.
Related: Prevalence rate, Incidence
Prevalence rate –
The proportion of persons in a population who have a particular
disease or attribute at a specified point in time or over a specified
period of time.
Prior authorization:
Requirement that a clinician obtain approval before a drug can
be dispensed and/or covered.
Probability  The
likelihood of an event occurring expressed as a number between
0 and 1. It is measured by the ratio of the event to the total
number of possible events. Example: the probability of flipping
a coin and coming up with heads is .5.
Proportion –
A type of ratio in which the numerator is included in the denominator.
The ratio of a part to the whole, expressed as a “decimal
fraction” (e.g., 0.2), as a fraction (1/5), or, loosely,
as a percentage (20%).
Related: Denominator, Numerator
Proxy outcome markers
– see Intermediate outcome markers.
Publication bias
– The bias toward the publishing of studies showing statistically
significant “positive” results.
Pvalue  Assuming there truly is no difference between the groups studied, the Pvalue is a calculated probability of observing a difference as big as or bigger than the one you observed in a study based on compatibility with an assumed standard distribution.
Problems with Pvalues:
 The Pvalue cannot tell you the chance the results are true or even how likely they are to be due to chance.
 “No test based on a theory of probability can by itself provide any valuable evidence of the truth or falsehood of a hypothesis.” [Neyman J, Pearson E. On the problem of the most ancient tests of statistical hypotheses. Philosophical Transactions of the Royal Society, Series A 1933; 231:289–337.]
 You do not know if the null hypothesis is true or not.
 You do not know if the sample is truly random and/or representative of the population.
 You do not know if the distribution in the population is standard.
 To reject a null hypothesis (e.g., no difference between groups), is not the same as accepting an opposite hypothesis (there is a true difference) as there could be other explanations for no difference (such as a bias or an insufficient number of people studied to show a true difference that actually exists, i.e., lack of power).
 The description of differences merely as “statistically significant” is not acceptable—precise Pvalues should be presented. [Numerous authors]
 Interpretation of confidence intervals in valid studies should focus on the range of values in the interval and the clinical importance of the results based on that range.
 A Pvalue of < 0.05, with a 95 per cent confidence interval which indicates that the true treatment effect may be close to the null value may be misleading. Approximately half of comparisons with P<0.05 are from null hypotheses which are true. [Sterne JA PMID: 11921006]
Therefore, the Pvalue has much more limited value than is frequently believed and isn’t very meaningful. One author likens it to a data compatibility issue: “The Pvalue is an indicator of the relative compatibility between the data and the null hypothesis, but it does not indicate whether the null hypothesis is a correct explanation for the data.” [Rothman KJ PMID: 2081237]
Consider the following scenario:
 Drug A compared to placebo
 100 patients in each group
 Mortality in Drug A group is 5%
 Mortality in placebo group is 15%
 Pvalue = .03
In this scenario, the Pvalue can be interpreted as follows:
“If it is true that there is no difference in mortality outcomes between Drug A and placebo, then there is a 3% statistical probability (chance) of observing a difference equal to or greater than the difference represented by 5 deaths out of 100 in the Drug A group as compared to 15 deaths out of 100 in the placebo group.”
Our approach:
In a study that is evaluated to be at low risk of bias, we may put greater reliance upon visual inspection of the differences between groups and examination of confidence interval, recognizing, however, that confidence intervals cannot be relied upon as meaning that a true answer lies within their range as there are related issues to Pvalues. Confirmatory studies and patterns provide us greater comfort with results.
Related: Statistical significance, Error (e.g., sampling error), Confidence intervals
Qualitative
research  Any research that utilizes nonquantitative methodology,
e.g., case studies or narrative description or interviews.
Quality indicator
– Oftentimes used interchangeably with “performance
measure.” Quality indicators are specific and measurable
elements of health care that can be used to assess the quality
of care.
Random
error – see Chance
Randomization –
Mechanism to assign, by chance, “intervention / no intervention”
assignments to study participants in randomized controlled trials.
 Mechanisms should
include random numbers which may be through a table or done
via computer.
 Simple randomization
is like a coin flip where each person has equal chance of being
assigned to either group.
 Blocked randomization
is used to create equally sized groups.
 Stratified randomization
randomizes subjects within selected criteria – such as
age or sex or anticipated confounders.
 Sequential methods
are more prone to bias especially when there has not been effective
concealment of allocation.
Randomized controlled
trial (RCT) – An experiment in which investigators control
an exposure or intervention, and subjects are randomly assigned
to the study group or to the control group or groups. Intervention
or exposure then is performed. Patients are followed to determine
outcomes such as improvements or harms.
Related: Experimental study
Random sample –
A sample derived by selecting individuals such that each individual
has the same probability of selection.
Rate – Derived
by dividing a numerator by a denominator. The numerator is a subset
of the denominator. Example: The percentage of diabetic eye exams
for Type I diabetics was 80% for our clinic.
Related: Denominator, numerator
Ratio – A numerator
and denominator (the numerator is not required to be a subset
of the denominator). Example –The ratio of women to men
in these studies was 1 to 5 (which can also be expressed as 1:5).
Or, the rate and ratio for diabetic eye exams in our clinic would
be 80/100.
Related: Denominator, numerator
Rebate: A cost offset
that pharmacy systems receive when purchasing large quantities
of drugs. AQ30Selection of drugs for large bulk purchase is based
on the drugs’ volume of use.
Recall bias –
Inaccuracies resulting from data collected from study participants
who are asked to retrospectively selfreport on study items of
interest. This kind of bias is most likely to occur when certain
behaviors are hard to track (e.g., diet) or are often hidden for
some reason such as being socially sensitive (e.g., sexual behaviors,
smoking behavior, etc.).
Related: Bias
Regression model
– A method for estimating the relationship of a dependent
variable to a variety of independent variables.
Regressiontothemean
– Returning to the “average” state. Meaning
that extreme test values are statistically likely to move to an
average over time. When patients present with extreme values and
then seem to have improvement, it may be falsely attributed to
an intervention when it is truly due to regressiontothemean.
A comparison group with no intervention can help expose this effect.
Relative risk (RR)
 A point estimate in which the risk of some healthrelated event,
such as disease or death, is compared between a study group and
a comparison group. Relative risk is expressed as the number of
times one group may be at risk over another. A relative risk of
less than 1 is a lower risk.
Formula: Risk in study group divided by risk in the comparison
group
Synonym: Risk ratio
Related: Measures of outcomes
Relative risk reduction
(RRR) – A point estimate in which the percent reduction
in events in the intervention or exposed group is compared to
the comparison group — relative risk reduction is the proportional
difference in size between outcomes. For example, if we had 15%
mortality in a comparison group as compared to 10% mortality in
the intervention group, since 10 is onethird smaller than 15,
the RRR would be 33%. Relative risk reduction begs the question
"relative to what?" Because RRR overestimates the actual
difference, it should always be used in conjunction with another
point estimate such as absolute risk reduction or numberneededtotreat
when communicating study results.
Formula: RRR = [((Comparison group outcomes  Intervention group
outcomes) / Comparison group outcomes) x 100] or 1Relative Risk
(RR)
Related: Measures of outcomes
Representative sample
– A sample whose characteristics correspond to those of
the original population or population of interest.
Retrospective study
– A study method to look at cause and effect after outcomes
have already occurred. Generally this refers to a casecontrol
study; however, cohort studies can use a retrospective method.
Risk – The
probability that an event will occur.
Risk adjustment –
The process of adjusting performance rates or other outcomes of
care to level the playing field due to differences in health status
between populations.
Risk factor –
An aspect of personal behavior or lifestyle, an environmental
exposure, or an inborn or inherited characteristic that is associated
with an increased occurrence of disease or other healthrelated
event or condition.
Risk ratio –
see Relative risk
Risk stratification
– The process of, or result of, separating a sample into
subsamples based on health status or risk factors such as age,
comorbidities, etc.
Safety
— Safety in research studies pertains to harms of various
interventions.
Related terms: Adverse events, ADEs, Harms, Risks, Side effect
Sample – A
selected subset of a population. A sample may be random or nonrandom
and it may be representative or nonrepresentative.
Related: Random sample, Representative sample
Sampling error 
See chance
Secondline therapy:
Therapy for patients failing on drugs established as “firstline”
therapy.
Selection bias –
See bias
Sensitivity (SN)
– Correct identification by a screening test or case definition
as having disease – of all those with a disease, the percent
testing positive (true positives). Sensitivity is derived from
calculations based on people who are known to have the condition.
Sensitivity is especially useful when it is important not to miss
a disease. High sensitivity is considered to "rule in"
a condition.
Formula = a / (a + c) from twobytwo table
Related: Measures of test function, Twobytwo table
Sensitivity analyses  Analyses undertaken to test the robustness of data (e.g., "what if" scenarios).
SIR  See Standardized incident ratio
Specificity (SP)
– Correct identification by a screening test or case definition
as not having disease – of all those without a disease,
the percent testing negative (true negatives). Specificity is
derived from calculations based on people who are known not to
have the condition. Specificity is especially useful when it is
important to avoid false positives. High specificity is considered
to "rule out" a condition.
Formula = d / (b + d) from twobytwo table
Related: Measures of test function, Twobytwo table
Standard deviation
(SD) – A widely used measure of dispersion of a frequency
distribution, equal to the positive square root of the variance.
Standard deviation can tell you how widely or how tightly variables
are distributed around the average. One standard deviation accounts
for approximately 68 percent of the distribution. Two standard
deviations account for approximately 95 percent.
Standardized incident ratio (SIR)  The ratio of observed occurrences to expected occurrences. Expected occurrences for SIR calculations are based on selected data sources e.g., the National Cancer Institute (NCI) Surveillance Epidemiology and End Results (SEER) database. For example, a SIR of 150 is interpreted as 50% more cases than the expected number; a SIR of 90 indicates 10% fewer cases than expected.
Standardized mortality ratio (SMR)  The ratio of observed deaths to expected deaths calculated using the expected rates (e.g., based on countryspecific age and sex matched general population data from the World Health Organization).
Statistical power
 The ability to reliably determine a statistically significant
relationship between interventions if one exists. In other words, power, by definition, means having sufficient people in a study to find a statistically significant difference between groups in an outcome of interest if such a difference truly exists. Therefore, any statistically significant outcome is, by definition, sufficiently powered for that outcome. Power comes
into play in planning and interpreting studies. Investigators
want to know the risk of drawing erroneous conclusion about efficacy
and so they do power calculations to estimate the number of subjects
needed in a study to show a statistically significant result if
one exists. It is conventional to accept a 5% risk of concluding
that an outcome is truly different in the study groups when they
are not (Type I or Alpha error). It is also conventional to accept
a 20% risk of concluding that an outcome is not statistically
significant when in truth there may be a true difference (Type
II or Beta error). Power can be determined by considering Alpha,
Beta, the event rate in the comparison group, and what is judged
to be a clinically significant difference between the groups.
As a practical matter, many users focus on the confidence intervals.
For valid studies, consider what you judge to be a reasonable
range for clinical significance – this need not be hard
and fast. For statistically significant findings, is the confidence
interval wholly within bounds for clinical significance? For nonsignificant
findings, is the confidence interval wholly beneath your limit
for clinical significance? A yes to these two questions means
likely conclusive findings for valid studies. No, means findings
are inconclusive.
Synonyms: Power, Power of a study
Related: Error
Statistical significance
– The extent to which study results are unlikely to be due
to chance. Expressed as a pvalue and it is typically set at <.05
due to convention. Statistical significance may be set lower,
but must be determined in advance of the study.
Related: Error, pvalue
Subgroup analysis
— Analysis of research results on subsets of patients in
a study, such as a subgroup of Caucasian women in a study of asthmatics
of all races, ages and genders. Subgroups for study must be determined
in advance of a study — otherwise the results are highly
likely to be due to chance.
Superiority trial
 Clinical trials which are conducted to determine if one intervention
is superior to another.
Surrogate end points
– See Intermediate outcome markers.
Survival analysis  a type of analysis that representsf the length of time to an outcome of interest, (e.g., timetopregnancy, timetocancer progression)
Synonyms: Life table analysis
Systematic review
– Studies that examine more than one study on a given topic
in a systematic way using predetermined criteria. The goal is
to provide a summary estimate of effect based on the scientific
weight of the studies.
Related: Metaanalysis.
Therapeutic substitution: Substitution of a drug with an agent
having a different chemical structure but similar clinical benefits.
Trend  A term used
to describe observed results in several studies which consistently
go in the same direction, but are not statistically significant.
Triangulation issues:
A catchall term for all the various considerations that need to
be made in making a decision.
Type 1 error  see
Error
Type 2 error  see
Error
TwobyTwo Table
— Table to report results and from which statistics are
calculated. See examples below:
Intervention
Example
Twobytwo
Table (2x2 Table) 
Not
Improved 
Improved 
Totals 
Intervention
Group 
a 
b 
a+b 
Control
Group 
c 
d 
c+d 
Totals 
a+c
= Total Not Improved 
b+d
= Total Improved 
a+b+c+d 
Diagnostic
Testing Example
Twobytwo
Table (2x2 Table) 
Condition
Present 
Condition
Absent 
Totals 
Test
Positive 
a
= true positives 
b
= false positives 
a+b 
Test
Negative 
c
= false negatives 
d
= true negatives 
c+d 
Totals 
a+c
= prevalence (in this population) 

a+b+c+d 
Validity – Closeness to truth. For example,
in performance measurement, the degree to which a measurement
actually measures or detects what it is supposed to measure.
Variable –
Any characteristic or attribute that can be measured.
Statistical tests
are chosen based on type of variables. The four main types of
variables are 
 Nominal (named
categories without any measurable scale such as ethnic groups)
 Dichotomous or
binary (two mutually exclusive categories resulting in “either
this or that” such as “death” or “survival”)
 Ordinal or ranked
(three or more variables that can be “ordered” or
ranked such as good/better/best or satisfied/neutral/unsatisfied)
 Continuous (can
be anywhere along a continuum, e.g., blood glucose readings)
Related: Dependent
variable, independent variable
Variance –
A measure of the spread of a variable about its mean value. In
a data set, a single point lies above or below the mean for the
entire dataset. A deviation score is the measure of how much each
point lies above or below the mean for the entire dataset. Variance
is the mean of all the deviation scores. It is a way of mathematically
stating how far deviations are from expected results.
Volunteer Bias  A bias that occurs when volunteers are enrolled in screening studies. Volunteers have been shown to have better outcomes than those who don’t (i.e., those who are recruited through active requests to participate), possibly due to the healthy user effect. In other words, patients who volunteer for screening may be “different” (healthier) from the larger population and may have a better prognosis.
Washout
period – Used in a crossover trial to discontinue patients
on a treatment before the second treatment is given to reduce
the likelihood that the first treatment will affect the outcome
even after it is stopped.
Withdrawal design  A study design in which an intervention to be tested is applied and then withdrawn to determine if any change reverts to preintervention levels. 