Glossary

A

 Absolute Risk Reduction (ARR)
The absolute arithmetic difference in the event rate between the control group (CER) and the treated (experimental) group (EER): ARR=CER-EER

Accuracy
The proportion of all tests, both positive and negative, that agreed with the ‘gold standard’ test.

Allocation concealment
Measure used to conceal allocation to study groups from those responsible for assessing patients for entry into the trial. Adequate allocation concealment: e.g.. Central randomisation; sequentially numbered; opaque; sealed envelopes; numbered or coded bottles or containers prepared by the pharmacy. Allocation not concealed: Inadequate or measures to conceal allocation from those responsible for entering patients into the trial. e.g.. Sealed envelopes that were not opaque; alternate patients, allocation by date of birth or hospital record number. Unclear allocation concealment: the authors did not report an adequate description of an allocation concealment approach that allowed classification

Applicability
How results from a clinical trial or meta-analysis apply to a specific individual or clinical setting. Patients’ individual characteristics may affect their outcome, making results from the literature more or less applicable to them.

B

 Baseline risk
The risk of an event occurring in an individual without the intervention. This is estimated from the event rate in the control group, but may differ in individual patients. People with a higher baseline risk may potentially benefit more from an intervention.

Bias
Bias results from factors other than the intervention systematically influencing the results of a study. Poor study design makes bias more likely; study techniques that limit bias include randomisation, stratification, intention-to-treat analysis, and the use of placebo and blinding. Bias can be unintentional, resulting from poor study design or execution. Bias can also be intentional, for example by use of an ineffective comparator drug when better comparators are available.

Blinded
Any or all of the clinicians, patients or participants, outcome assessors or statisticians were unaware of who received which study intervention. Blinding may be considered unclear if the authors did not adequately describe whether the study was blinded.

C

 Case control study
Involves identifying patients who have the outcome of interest (cases) and control patients without the same outcome (controls) and looking back to see if they had the exposure of interest. It may be the only ethical way of studying some questions of Harm.

Case report
A clinical report on a single patient with a disease or outcome of interest, with or without an intervention.

Case series
A report on a series of patients with a disease, treatment, or outcome of interest. No control group is involved.

CER
Control event rate: the event rate in the comparator arm (See Event Rate)

Clinical Practice Guideline
A systematically developed statement designed to assist practitioners and patients make decisions about appropriate health care for specific clinical circumstances. Clinical practice guidelines are often produced by expert groups or working parties and should include transparent information about how recommendations were developed. They should be produced with knowledge of local conditions and can provide useful practical guidance for clinicians.

Clinical versus Statistical significance
Study findings that may be only modestly important for patients may be statistically significant if the study is large enough. It is important to ask whether results of a statistically significant study are actually clinically significant. Conversely, a small study may be underpowered to detect a statistically significant difference, even if the differences found would be very clinically important. In this setting, more studies or a meta-analysis are needed.

Cluster randomisation
Cluster randomisation is a way of randomising groups of participants to the same intervention as a group, rather than as individuals. For example, people in the same school or hospital may be randomised together to receive the same intervention, even if the intervention is delivered at the level of the individual. Results should be analysed as groups rather than as individuals.

Cochrane Collaboration
The Cochrane Collaboration is a global network of volunteer reviewers, supported by methodological experts and resources, dedicated to performing high quality systematic reviews of health care topics which are globally accessible to health care professionals and consumers.

Cohort study
This study design is also called a ‘prospective observational study’. Groups of patients, called ‘cohorts’, are identified and followed over time for the outcomes of interest.

Confidence Interval
Quantifies the uncertainty of a measurement. It is usually reported as the 95% CI, which is the range of values within which we can be 95% sure that the true value for the whole study population lies. I.e.. if the study were repeated 100 times, the study results for 95 of those studies would fall within this interval.

Conflict of Interest
A conflict of interest occurs where people involved in the conduct or reporting of a study also have financial or other interests in the outcome of the study. A conflict of interest does not necessarily mean that the results are any less valid; however, conflict of interest should be reported rather than hidden. Published manuscripts usually contain a Conflict of Interest statement from authors.

Confounding variable
A variable which is not the intervention or exposure being studied, but which distorts the measure of effect of the intervention or exposure under study because it is associated with that intervention/exposure. For example, results of an RCT may be confounded because one treatment requires regular interaction with nursing or medical staff, while another treatment can be self-administered at home. In this situation, a study should be designed to give all treatment groups the same interaction with staff to minimise confounding.

CONSORT
CONsolidated Standards of Reporting Trials – the CONSORT group developed the CONSORT statement to improve the reporting of randomised clinical trials, with a checklist and flow diagram to enable readers to understand the trial’s design, conduct, analysis and interpretation in order to assess the validity of the trial.

Cost-Benefit Analysis
Converts effects into the same monetary terms as the costs to assess whether the cost of an intervention is worth the benefit.

Cost-Effectiveness Analysis
Converts effects into the same monetary terms as the costs to assess whether the cost of an intervention is worth the benefit, and additionally includes the net cost of providing the intervention.

Cost-minimisation analysis
If two interventions are known to be equal in efficacy and toxicity, only costs are analysed and the least costly alternative is chosen

Critical appraisal
The systematic and explicit process of interpreting clinical research evidence for validity and relevance.

Critically appraised topic (CAT)
A short summary of a search and critical appraisal of the literature created to answer a specific clinical question, for the purpose of making clinical decisions.

Cross-Sectional Study Design
Cross-sectional studies are observational studies that take place at a single point in time or over a set time interval. The exposure and outcome are determined simultaneously and provide a ‘snapshot’.

Crossover Study Design
A study design in which the same group of patients each experiences all of two or more experimental therapies one after the other. The order of administration may be random or pre-specified. Investigators need to pay careful attention for the possibility of an effect of treatment order or a carry-over effect. Carry-over effects can be minimised by a wash-out period between treatments, or by delaying the start of outcome measurement for the second treatment.

D

Decision analysis
The use of quantitative methods to analyse decisions under conditions of uncertainty, with outcomes represented by probabilities.

Double blind
At least two groups (e.g.. Patient and physician) don’t know which treatment the patient has been allocated to. This methodology helps prevent bias in clinical research.

E

EBM
Evidence Based Medicine

Ecological Survey
A survey based on pooled data for a particular population at some point or points in time, to investigate the relationship of an exposure to a known or presumed risk factor for a specified outcome. Ecological surveys can be useful in determining whether the widespread uptake of new clinical knowledge has changed population outcomes. e.g.. studying whether outcomes for patients with early breast cancer have improved since mammography was introduced.

Effectiveness
The benefit of an intervention under real-life conditions for its use. I.e.. once the treatment is used in patients who are less well-selected, or by doctors who are less well-trained in dose optimisation, the treatment may be less effective.

Efficacy
The benefit of an intervention under ideal conditions for its use.

Equivalence trials
Equivalence trials are used to exclude a meaningful difference between two interventions. Finding no difference between two interventions in a randomised superiority trial is not the same as demonstrating that the two interventions are equivalent.

Event
The occurrence of a study outcome that is either present or not. For example, a patient is either dead or alive, has had disease progression or has not.

Event Rate
The proportion of patients in a group who experience the event, or outcome of interest. If an event is observed in 15 of 100 patients, the event rate is 0.15.

Evidence based Medicine (EBM)
Evidence based medicine is the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients. The practice of evidence based medicine means integrating individual clinical expertise with the best available external clinical evidence from systematic research (Sackett DL et al. BMJ 1996;312:71-72 ).

Exclusion criteria
Patient characteristics which exclude that individual from participating in a clinical trial.

F

Factorial design
A clinical trial design which evaluates more than one intervention in a single clinical trial, by randomising patients independently to two or more different intervention and control arms. Ideally, there will be no interaction between the effects of the different interventions. A factorial design study should be analysed for each intervention independently, rather than for the individual combinations of interventions. For example, if the clinical trial has a 2 x 2 factorial design randomising patients to chemotherapy or no chemotherapy, and radiotherapy or no radiotherapy, the four possible groups are chemotherapy alone, radiotherapy alone, both treatments, or neither treatment. However, only the effects of chemotherapy versus none, or radiotherapy versus none, should be analysed.

False negative rate
The rate at which a test or examination is negative, when the condition is actually present. I.e.. the test fails to detect the condition. As the sensitivity of a test increases, the false negative rate decreases. False negative rate = 1-sensitivity.

False positive rate
The proportion of positive tests amongst patients who do not have the condition. I.e.. the test falsely suggests that the condition is present. As the specificity of a test increases, the false positive rate decreases. False positive rate = 1-specificity

Forest plot
A Forest plot is a graphical display of the results of multiple studies, including a statistical synthesis of the pooled study results. Results of individual studies are shown, usually with a square representing the individual study and size, and a horizontal line representing the confidence intervals. A vertical line shows the line of no effect, and individual study results may lie to the left or right of the vertical line depending on whether the result favoured the intervention or comparator arm.

Funnel plot
A Funnel plot is a graphical display of the results of multiple studies that is used to identify whether the results have been affected by publication bias. It is a scatter plot of treatment effect versus study size. A symmetrical funnel plot suggests that there is no publication bias. An asymmetrical funnel may suggest systematic lack of reporting of small, usually negative, studies.

G

Gold standard
The reference test, which should be the best previously available, against which a new test is evaluated. The ‘gold standard’ is assumed to have 100% sensitivity and specificity for the diagnosis in question for the purposes of the study.

H

Heterogeneity
A statistical test to determine whether clinical trials included in a meta-analysis are sufficiently similar for their results to be pooled statistically. I.e.. is the meta-analysis combining apples with apples, or apples with oranges?

I

ICMJE
International Committee of Medical Journal Editors.

Incidence
The rate at which an event, or illness, occurs in a given population over time.

Inclusion criteria
Patient characteristics which are required to allow that individual to participate in a clinical trial.

Individual patient data
Raw patient data from individual clinical trials for use in meta-analysis (as opposed to use of the published data). This requires the co-operation of individuals or organisations who performed the original studies to release data.

Intention to Treat (ITT) Analysis
When outcomes in a randomised clinical trial are analysed according to the treatment group the patient was originally allocated to. This means that patients who were randomised to Arm A but never received that treatment are treated in the analysis as if they did participate in Arm A. ITT analysis is ‘best practice’ for efficacy outcomes in clinical trials and minimises the risk of bias in favour of either treatment. However, toxicities should be analysed by treatment received.

L

Levels of evidence
Assigning levels of evidence is a way of ranking study designs according to how likely they are to be free from systematic bias.

Likelihood Ratio (LR)
Likelihood ratios describe how well a test rules in or rules out a disorder; i.e., the likelihood that a given test result would be expected in a patient with the target disorder compared to the likelihood that the same result would be expected in a patient without that disorder. The LR for a positive test (positive likelihood ratio) = sensitivity / (1 – specificity). The LR for a negative test (negative likelihood ratio) = (1 – sensitivity) / specificity. A test with a LR of 1.0 does not change the probability of disease. For a positive test, as the LR increases, the better it ‘rules in’ the disease; a LR>10 would be considered good. For a negative test, as the LR decreases, the better it ‘rules out’ the disease; in this setting, a LR < 0.1 would be considered good.

M

MeSH
Medical Subject Headings: a thesaurus of medical terms used by many databases and libraries (notably Medline) to index and classify medical information

Meta-analysis
An review of a focused clinical question which uses quantitative methods to summarise the results of a number of independent trials addressing the same question. Ideally a meta-analysis should be performed in the context of a systematic review and critical appraisal of the studies.

Morbidity
Rate of illness without death

Mortality
Rate of death

N

N-of-1 Trials
The patient undergoes pairs of treatment periods organised so that one period uses the experimental treatment and one period involves the use of an alternate or placebo therapy. The patient and physician are blinded, if possible, and the outcomes are monitored. Treatment periods are replicated until the clinician and patient are convinced that the treatments are definitely different or not different.

Negative Predictive Value
The proportion of people with a negative test who are free of disease.

Null hypothesis
The null hypothesis assumes that there is no difference between study groups, and that the intervention you are studying has no effect. Any difference in outcome disproves the null hypothesis.

Number Needed to Harm (NNH)
The number of patients who, if they received the experimental treatment, would lead to one additional person being harmed compared to patients who receive the control treatment.

Number Needed to Treat (NNT)
The number of patients who need to be treated to prevent one bad outcome. It is the inverse of the ARR: NNT = 1 / ARR, rounded up to the next whole number.

O

Observational study
A non-experimental study design in which the exposure is not assigned by the investigators. For example, case-control and cohort studies are observational studies.

Odds
A ratio of events to non-events. I.e.. EER/CER. If the event rate for a disease is 0.1 (10%), it’s non-event rate is 0.9 (90%), and therefore its odds are 1:9, or 0.111.

Odds Ratio
The odds ratio is a measure of the size of the association between an exposure and an outcome (i.e.. disease, death). Observational studies may report results as odds ratios or relative risks. Odds ratios are used to estimate relative risk in case-control studies. Odds ratios are also used to describe the odds of an experimental patient experiencing an adverse event relative to a control patient.

Overview
A systematic review and summary of the literature using EBM principles.

P

p value
The probability that a particular result would have happened by chance. For example, a p value of 0.01 tells us that if the study were done 100 times, the only once would these results have been reproduced by chance alone. A p value of 0.05 is often considered statistically significant, meaning that the results would have happened by chance alone 5 times in 100 similar studies. The use of p<0.05 to represent a statistically significant result is historical convention.

Per protocol analysis
68Analysis of outcomes only for patients who complete the study, or analysed for treatment actually received. Unlike ‘Intention to treat analysis’, this practice may bias results in favour of the active treatment.

PICO
Acronym for the four components of a well-built clinical question (focused clinical question), also referred to as a ‘PICO question’. Population, Intervention, Comparator, Outcomes.

Placebo
A substance or sham treatment given to the control group of a clinical trial. The placebo should be identical in appearance and taste to the experimental treatment, but should lack any activity against the disease, or any side effects.

Population
The target group (e.g. healthy males 17 – 60)

Positive Predictive Value
The proportion of people with a positive test who have the disease in question. Although the sensitivity and specificity of a test always remain the same, the PPV increases as the probability of disease changes in a population or in an individual patient. Two tests with similar high sensitivity and specificity will have very different PPV if one is used to diagnose a rare disease, and the other to diagnose a common disease. The PPV of a screening test may be increased by restricting its use to those at high risk of the disorder.

Post-test probability
The likelihood the patient has the condition you are testing for after the test is performed.

Power
The chance that a study will correctly detect a statistically significant difference between study groups, if a true difference is there. The power of a study is described as the β of a study. I.e.. if β then the chance of a study detecting a statistically significant difference is 90%. A larger study usually has more power to detect a given difference.

Pre-test probability
An estimate of how likely it is that the patient has the disease, before the test is performed. Clinical history and examination help a clinician to make this estimate based on the patient’s symptoms and signs. This process is usually implicit, but can be made more explicit by use of information from epidemiological studies (e.g.. cardiovascular risk calculators).

Prevalence
The proportion of people in a defined population who have the disorder or condition of interest. In a question of Diagnosis, this will be the same as the ‘pre-test probability’ in that population; however an individual patient may have characteristics that make their pre-test probability different from the population prevalence of the disorder.

Primary endpoint
The most important outcome that the investigators are measuring. The sample size and statistical calculations for the trial should be based on the primary endpoint.

Prospective study
A prospective study starts with patient selection and is done forwards in time. This design minimises hidden bias as compared with a retrospective study design.

Prospective, blind comparison to a gold standard
These are studies that show the efficacy of a diagnostic test. All patients with varying degrees of an illness, or clinical suspicion of an illness, and who are eligible for an investigation or clinical test, undergo both investigations or tests – the test under investigation and the ‘gold standard’ test. The ‘gold standard’ test is the current most sensitive/specific for the condition. The investigators interpreting the test results are blinded to the results of the alternative test.

Publication bias
Bias caused by omission of some trials from a systematic review or meta-analysis because they have not been published, or have been published in the non-English literature. Inconclusive trials, trials with negative results, or those which poor methodology are less likely to be published. This may lead to an over-estimation of the treatment effect if only fully published trials are considered. Authors of systematic reviews may identify unpublished trials and contact the investigators for results, in an effort to minimise publication bias.

Q

Quality Adjusted Life Year (QALY)
A method for representing the concept of both quality and quantity of life in one figure. A year of perfect health is worth 1.0 QALY. Years of less than perfect health are worth less than 1.0 QALY. Thus an intervention may gain years of life, but at a cost of toxicities for some of that time (thus making those years where toxicity or less than perfect health was experienced worth less).

Quality of Life (QOL)
A multi-dimensional concept that includes aspects of an individual’s health, social, emotional, physical, and functional abilities and comfort. Quality of Life, and in particular Health-related QOL, can be measured by validated questionnaires and is an important outcome in many clinical trials.

R

Randomised Controlled Clinical Trial
A group of patients is randomly allocated into either the control group or the experimental group. These groups are followed up for the variables/outcomes of interest. Randomisation should ensure that baseline characteristics are similar between the different groups, and helps minimise bias.

Relative Risk
The risk of developing an event in the experimental (or exposed) group (EER), divided by the risk of developing an event in the control (unexposed) group (CER): RR = EER/CER

Relative Risk Reduction (RRR)
The percent reduction in events in the treated group event rate (EER) compared to the control group event rate (CER): RRR = (CER-EER) / CER * 100. RRR should be supplemented by an estimate of absolute risk reduction, which is more meaningful to an individual.

Retrospective study
A study in which the outcomes of interest have occurred before the study has begun.

Risk Ratio
The ratio of risk in the treated group (EER) to the risk in the control group (CER): RR =EER/CER Risk ratio is used in randomised trials and cohort studies

Run-in period
A period before a study starts, either before the intervention is used or using the intervention to identify compliant or responsive patients. A run-in period may be appropriately used, for example to wean patients from another medication. However, when used to select responding, tolerant, or compliant patients it may bias the study in favour of the intervention. At times (e.g.. with targeted therapies for cancer) this may also be appropriate if the biological target of the treatment is unknown, and only a subset of the whole population is anticipated to benefit. In this setting, the trial may be used to molecularly characterise patients who may respond to treatment. However, results should not then be extrapolated to the population as a whole.

S

Sample size
The number of patients in a study. A study with a large sample size may be able to find small differences between interventions that may or may not be clinically significant. A study with a small sample size may be ‘underpowered’ to detect even a true difference between groups.

Selection bias
A systematic difference between the characteristics of the comparison and intervention groups in a clinical trial. Selection bias can be minimised by randomisation. Selection bias can also occur in a systematic review through reviewer decisions to include or exclude studies in the review. ‘Selection bias’ may also refer to the selection of a group of patients for participation in a clinical trial, through the explicit or implicit use of exclusion and inclusion criteria, who do not truly represent the population to whom the results will be generalised.

Sensitivity
The probability that a person with the disease or condition will have a positive test. A high sensitivity is most valuable.

SnNout
when a sign/test has a high sensitivity, a negative result rules out the diagnosis.

Specificity
The probability that a person without the disease or condition will have a negative test. A high specificity is most valuable.

SpPin
when a sign/test has a high specificity, a positive result rules in the diagnosis.

Statistical significance
A measure of the confidence that any difference observed between the study groups is due to the study intervention rather than chance alone. Statistical significance is not the same as clinical significance.

Stratification
A technique for ensuring that the different arms of a clinical trial are balanced for important prognostic factors which could otherwise bias the study outcome. Stratification may be performed for one or more factors. For example, if randomisation is stratified for study site, each site should have equal numbers of patients randomised to each treatment arm. Stratification for many factors can only be performed if the sample size is large.

Surrogate endpoint
An outcome measure that is not directly related to the key clinical outcome, but is believed to represent the outcome of interest. For example, many surrogate endpoints are changes in biomarkers or physiological measures, such as normalisation of thyroid function tests, or control of hypertension. Surrogate endpoints should be validated to represent the outcome of interest, and are most useful when the key clinical outcome requires long follow-up.

Survival curve
A plot of the probability that a patient in a particular group will be event-free at a certain time point. Follow-up time appears on the x-axis, starting from time 0 for all patients. The y-axis shows a probability from 0.0 to 1.0. Survival curves can be applied to other endpoints, not just mortality. Patients who are still alive (or have not had the outcome of interest) at a certain time point are usually ‘censored’, and can be represented as a vertical dash on the curve.

Systematic review
A systematic review should ask a focused clinical question, describe a search strategy and inclusion and exclusion criteria for the review, and systematically search for all relevant studies in the literature. Studies should then be appraised for quality by pre-determined criteria in order to answer the question based on the best available evidence. A systematic review may or may not include a meta-analysis for statistical synthesis of the study results.

T

Type I error
Also known as alpha. The probability of concluding that the null hypothesis is false, when in fact it is true. Typically set at 0.01 or 0.05. i.e. concluding that an ineffective treatment is actually effective.

Type II error
Also known as beta (β). The probability of concluding that the alternative hypothesis is false, when in fact it is true. I.e.. concluding that an effective treatment is actually ineffective. Power = 1-β. Often set at 0.1 or 0.2.

U

Unblinded
All participants in the trial (clinicians, patients, outcome assessors and statisticians) were aware of who received which study intervention.

V

Validity
The extent to which study results are likely to be true and free of bias. Internal validity can be assessed by determining whether the study design was likely to be free from bias. External validity can be assessed by determining whether the results of the study can be applied to patient populations which were not on the clinical trial.