Absolute risk measures the size of a risk in a person or group of people. This could be the risk of developing a disease over a certain period, or it could be a measure of the effect of a treatment – for example, how much the risk is reduced by treatment in a person or group.
There are different ways of expressing absolute risk. For example, someone with a 1 in 10 risk of developing a certain disease has “a 10% risk” or “a 0.1 risk”, depending on whether percentages or decimals are used.
Absolute risk doesn’t compare changes in risk between groups – for example, risk changes in a treated group compared to risk changes in an untreated group. That’s the function of relative risk.
Before and after study
A before and after study measures particular characteristics of a population or group of individuals at the end of an event or intervention, and compares them with those characteristics before the event or intervention. The study gauges the effects of the event or intervention.
Blinding is not telling someone what treatment a person has received or, in some cases, the outcome of their treatment. This is to avoid them being influenced by this knowledge. The person who’s blinded could be either the person being treated or the researcher assessing the effect of the treatment (single blind), or both of these people (double blind).
A case-control study is an epidemiological study that’s often used to identify risk factors for a medical condition. This type of study compares a group of patients who have that condition with a group of patients that don’t, and looks back in time to see how the characteristics of the 2 groups differ.
Case crossover studies
Case crossover studies look at the effects of factors thought to increase the risk of a particular outcome in the short term. For example, this type of study might be used to look at the effects of changes in air pollution levels on the short-term risk of asthma attacks. Individuals who have had the outcome of interest are identified and act as their own control.
The presence or absence of the risk factor is assessed for the period immediately before the individual experienced the outcome. This is compared with the presence or absence of the risk factor when the individual didn’t experience the outcome (control period). If there’s a link between the risk factor and the outcome, it would be expected to have been present in the period just before the outcome more often than in the control period.
A case series is a descriptive study of a group of people, who usually receive the same treatment or have the same disease. This type of study can describe characteristics or outcomes in a particular group of people, but can’t determine how they compare with people who are treated differently or who don’t have the condition.
Clinical practice guidelines
Clinical practice guidelines are statements that are developed to help practitioners and patients make decisions about the appropriate healthcare for specific clinical circumstances.
Cluster randomised controlled trial
In a cluster randomised controlled trial, people are randomised in groups (clusters) rather than individually. Examples of clusters that could be used include schools, neighbourhoods or GP surgeries.
This study identifies a group of people and follows them over a period of time to see how their exposures affect their outcomes. This type of study is normally used to look at the effect of suspected risk factors that can’t be controlled experimentally – for example, the effect of smoking on lung cancer.
A confidence interval (CI) expresses the precision of an estimate and is often presented alongside the results of a study (usually the 95% confidence interval). The CI shows the range within which we’re confident that the true result from a population will lie 95% of the time.
The narrower the interval, the more precise the estimate. There’s bound to be some uncertainty in estimates because studies are conducted on samples and not entire populations.
By convention, 95% certainty is considered high enough for researchers to draw conclusions that can be generalised from samples to populations. If we’re comparing 2 groups using relative measures, such as relative risks or odds ratios, and see that the 95% CI includes the value of one in its range, we can say there’s no difference between the groups.
This confidence interval tells us that, at least some of the time, the ratio of effects between the groups is one. Similarly, if an absolute measure of effect, such as a difference in means between groups, has a 95% CI that includes 0 in its range, we can conclude there’s no difference between the groups.
Confounding factor (confounder)
A confounder can distort the true relationship between two (or more) characteristics. When it isn’t taken into account, false conclusions can be drawn about associations. An example is to conclude that if people who carry a lighter are more likely to develop lung cancer, it’s because carrying a lighter causes lung cancer. In fact, smoking is a confounder here. People who carry a lighter are more likely to be smokers, and smokers are more likely to develop lung cancer.
A control group (of cells, individuals or centres, for example) serves as a basis of comparison in a study. In this group, no experimental stimulus is received.
This is an epidemiological study that describes characteristics of a population. It’s “cross-sectional” because data is collected at one point in time and the relationships between characteristics are considered. Importantly, because this study doesn’t look at time trends, it can’t establish what causes what.
A diagnostic study tests a new diagnostic method to see if it’s as good as the “gold standard” method of diagnosing a disease. The diagnostic method may be used when people are suspected of having a disease because of signs and symptoms, or to try to detect a disease before any symptoms have developed (a screening method).
In ecological studies, the unit of observation is the population or community. Common types of ecological study are geographical comparisons, time trend analysis, or studies of migration.
Epidemiology is the study of factors that affect the health and illness of populations.
An experiment is any study in which the conditions are under the direct control of the researcher. This usually involves giving a group of people an intervention that wouldn’t have occurred naturally. Experiments are often used to test the effects of a treatment in people, and usually involve comparison with a group who don’t get the treatment.
Gene expression is a term used to describe the influence the “information” contained in genes can have on a cellular level – in most cases, in terms of the way specific proteins are created.
Genome-wide association study
This study looks across the entire genetic sequence (genome) to identify variations in this sequence that are more common in people with a particular characteristic or condition and may be involved in producing that characteristic or condition.
A measure of the relative probability of an event in 2 groups over time.
It’s similar to a relative risk, but takes into account the fact that once people have certain types of event, such as death, they’re no longer at risk of that event.
A hazard ratio of 1 indicates that the relative probability of the event in the 2 groups over time is the same. A hazard ratio of more than or less than 1 indicates that the relative probability of the event over time is greater in one of the two groups.
If the confidence interval around a hazard ratio doesn’t include 1, the difference between the groups is considered to be statistically significant.
Intention-to-treat (ITT) analysis is the preferable way to look at the results of randomised controlled trials (RCTs).
In ITT analysis, people are analysed in the treatment groups to which they were assigned at the start of the RCT, regardless of whether they drop out of the trial, don’t attend follow-up, or switch treatment groups.
If follow-up data isn’t available for a participant in one of the treatment groups, the person would normally be assumed to have had no response to treatment, and that their outcomes are no different from what they were at the start of the trial.
This helps make sure RCTs don’t show that a particular treatment being tested is more effective than it actually is. For example, if 50 people were allocated to the treatment group of an RCT, perhaps 10 might drop out because they got no benefit.
If all 50 were analysed by ITT analysis, with 10 assumed to have had no benefit, this gives a more reliable indication of the effect of the treatment than just analysing the remaining 40 people who stayed on treatment because they felt they were getting the benefit.
Levels of evidence
This is a hierarchical categorisation (ranking) of different types of clinical evidence. It’s partly based on the type of study involved, and ranks evidence according to its ability to avoid various biases in medical research.
Several ranking schemes exist that are specific to the question posed in the research. Studies with the highest ranking are those that provide the best evidence that a result is true.
Examples of studies ranked in order from high-level to low-level evidence are:
• systematic reviews
• single randomised controlled trials
• controlled trials without randomisation
• prospective cohort studies
• case-control studies
• cross-sectional studies
• case series
• single case reports
The expert opinions of respected authorities – based on clinical experience, descriptive studies, physiology, bench research or first principles – are often thought of as the lowest level evidence.
Although there are different systems, some of which take into account other aspects of quality including the directness of the research, the levels are designed to guide users of clinical research information as to which studies are likely to be the most valid.
A Likert scale is a commonly used rating scale that measures attitudes or feelings on a continuous linear scale, usually from a minimum “strongly agree” response to a maximum “strongly disagree” response, or similar. Likert scales can be 5-point, 6-point, 10-point etc depending on the number of response options available.
A longitudinal study is one that studies a group of people over time.
This is a mathematical technique that combines the results of individual studies to arrive at one overall measure of the effect of a treatment.
A narrative review discusses and summarises the literature on a particular topic, without generating any pooled summary figures through meta-analysis. This type of review usually gives a comprehensive overview of a topic, rather than addressing a specific question, such as how effective a treatment is for a particular condition. Narrative reviews don’t often report on how the search for literature was carried out or how it was decided which studies were relevant to include. Therefore, they’re not classified as systematic reviews.
Negative predictive value
This is one of a set of measures used to show the accuracy of a diagnostic test (see sensitivity, specificity and positive predictive value). The negative predictive value (NPV) of a test is a measure of how accurate a negative result on that test is at identifying that a person doesn’t have a disease. The NPV is the proportion of people with a negative test result who don’t truly have a disease.
For example, if a test has an NPV of 75%, this means that 75% of the people who test negative are truly disease free, while 25% who test negative have the disease (false negatives). The NPV for a test varies depending on how common the disease is in the population being tested. An NPV is usually lower (false negatives are more common) when disease prevalence is higher.
Nested case-control study
A nested case-control study is a special type of case-control study in which “cases” of a disease are drawn for the same cohort (population of people) as the controls to whom they’re compared. These studies are sometimes called case-control studies nested in a cohort or case-cohort studies. The collection of data on the cases and controls is defined before the study begins.
Compared with a simple case-control study, the nested case-control study can reduce recall bias (where a participant remembers a past event inaccurately) and temporal ambiguity (where it’s unclear whether a hypothesised cause preceded an outcome).
It can be less expensive and time consuming than a cohort study. Incidence and prevalence rates of a disease can sometimes be estimated from a nested case-control cohort study, whereas they can’t from a simple case-control study, as the total number of exposed people (the denominator) and the follow-up time aren’t usually known.
In this type of study, participants aren’t randomly allocated to receiving (or not receiving) an intervention.
In an observational study, researchers have no control over exposures and instead observe what happens to groups of people.
An odds ratio is one of several ways to summarise the association between an exposure and an outcome, such as a disease. Another commonly used approach is to calculate relative risks.
Odds ratios compare the odds of the outcome in an exposed group with the odds of the same outcome in an unexposed group. Odds tell us how likely it is an event will occur, compared with the likelihood that the event won’t happen. Odds of 1:3 that an event occurs, such as a horse winning a race, means the horse will win once and lose 3 times (over 4 races). Odds ratios are a way of comparing events across groups who are exposed and those who aren’t.
Open access means that a study or article is available free of charge, usually online. To access full articles in most medical journals you usually have to pay a subscription or make a one-off payment (these types of articles are often referred to as paywalled content).
Some fully open access journals are funded by non-profit organisations. Others meet their running costs by charging individual authors a fee for publication.
Occasionally, a paywalled journal will release individual articles on an open access basis (often those with important public health implications).
Open label means that investigators and participants in a randomised controlled trial are aware of what treatment is being given and received (the study isn’t blinded).
Peer review involves giving a scientific paper to one or more experts in that field of research to ask whether they think it’s of good enough quality to be published in a scientific journal. Studies that aren’t of sufficient quality won’t be published if their faults aren’t corrected. Journals that use peer review are considered to be of better quality than those that don’t.
Per-protocol analysis, sometimes called on-treatment analysis, is one way to analyse the results of randomised controlled trials (RCTs). It analyses the outcomes of only the participants who receive a trial treatment exactly as planned, and excludes participants who don’t.
This approach can exclude participants who drop out of the trial for important reasons (for example, because the treatment isn’t working for them or they experience side effects). Excluding these people from the analysis can bias the results, making the treatment look better that it would be in a real-world situation where some people may not follow the treatment plan perfectly.
Per-protocol analysis can give a good estimate of the best possible outcome of treatment in those who take it as intended. Intention-to-treat (ITT) analysis is the alternative, and generally preferable, way to look at the results of RCTs because it gives a better idea of the real-world effects of treatment.
Person years describes the accumulated amount of time that all the people in the study were being followed up. So, if 5 people were followed up for 10 years each, this would be equivalent to 50 person years of follow-up.
Sometimes the rate of an event in a study is given per person year rather than as a simple proportion of people affected to take into account the fact that different people in the study may have been followed up for different lengths of time.
Phase I trials
Phase I trials are the early phases of drug testing in humans. These are usually quite small studies that primarily test the drug’s safety and suitability for use in humans, rather than its effectiveness.
They often involve between 20 and 100 healthy volunteers, although they sometimes involve people who have the condition the drug is aimed at treating. To test the drug’s safe dosage range, very small doses are given initially and are gradually increased until the levels suitable for use in humans are found.
These studies also test how the drug behaves in the body, examining how it’s absorbed, where it’s distributed, how it leaves the body, and how long it takes to do this.
Phase II trials
During this phase of testing, a drug’s effectiveness in treating the targeted disease in humans is examined for the first time and more is learnt about appropriate dosage levels.
This stage usually involves 200 to 400 volunteers who have the disease or condition the drug is designed to treat. The drug’s effectiveness is examined, and more safety testing and monitoring of its side effects are carried out.
Phase III trials
In this phase of human testing of treatments, the effectiveness and safety of the drug undergoes a rigorous examination in a large, carefully controlled trial to see how well it works and how safe it is.
The drug is tested in a much larger sample of people with the disease or condition than before, with some trials including thousands of volunteers. Participants are followed up for longer than in previous phases, sometimes over several years.
These controlled tests usually compare the new drug’s effectiveness with either existing drugs or a placebo. These trials are designed to give the drug as unbiased a test as possible to ensure that the results accurately represent its benefits and risks.
The large number of participants and the extended period of follow-up give a more reliable indication of whether the drug will work, and allows rarer or longer term side effects to be identified.
Positive predictive value
This is one of a set of measures used to show how accurate a diagnostic test is (see sensitivity, specificity and negative predictive value).
The positive predictive value (PPV) of a test is how well the test identifies people who have a disease. The PPV is the proportion of people with a positive test result who truly have the disease. For example, if a test has a PPV of 99%, this means 99% of the people who test positive will have the disease, while 1% of those who test positive won’t (false positives).
The PPV of a test varies depending on how common the disease is in the population being tested. A test’s PPV tends to be higher in populations where the disease is more common and lower in populations where the disease is less common.
These are in vitro (for example, in cell cultures) and in vivo laboratory animal tests on drugs in development carried out to ensure they’re safe and effective before they go on to be tested in humans (clinical studies).
Prevalence describes how common a particular characteristic (for example, a disease) is in a specific group of people or population at a particular time. Prevalence is usually assessed using a cross-sectional study.
Prospective observational study
This study identifies a group of people and follows them over a period of time to see how their exposures affect their outcomes. A prospective observational study is normally used to look at the effect of suspected risk factors that can’t be controlled experimentally, such as the effect of smoking on lung cancer.
A prospective study asks a specific study question (usually about how a particular exposure affects an outcome), recruits appropriate participants, and looks at the exposures and outcomes of interest in these people over the following months or years.
Publication bias arises because researchers and editors tend to handle positive experimental results differently from negative or inconclusive results. It’s especially important to detect publication bias in studies that pool the results of several trials.
Qualitative research uses individual in-depth interviews, focus groups or questionnaires to collect, analyse and interpret data on what people do and say. It reports on the meanings, concepts, definitions, characteristics, metaphors, symbols and descriptions of things. It’s more subjective than quantitative research, and is often exploratory and open-ended. The interviews and focus groups involve relatively small numbers of people.
Quantitative research uses statistical methods to count and measure outcomes from a study. The outcomes are usually objective and predetermined. A large number of participants are usually involved to ensure the results are statistically significant.
Randomised controlled trial (RCT)
This is a study where people are randomly allocated to receive (or not receive) a particular intervention (this could be 2 different treatments or 1 treatment and a placebo). This is the best type of study design to determine whether a treatment is effective.
Randomised crossover trial
This is a study in which people receive all of the treatments and controls being tested in a random order. This means that people receive one treatment, the effect of which is measured, and then “cross over” into the other treatment group, where the effect of the second treatment (or control) is measured.
Recall bias is when a person’s recall of their exposure to a suspected disease risk factor could be influenced by the knowledge that they’re now suffering from that particular disease. For example, someone who’s suffered a heart attack may recall having a highly stressed job. The stress they now report experiencing may be subtly different from the stress they would have reported at the time, before they developed the disease.
Relative risk compares a risk in 2 different groups of people. All sorts of groups are compared to others in medical research to see if belonging to a particular group increases or decreases the risk of developing certain diseases. This measure of risk is often expressed as a percentage increase or decrease, for example, “a 20% increase in risk” of treatment A compared with treatment B. If the relative risk is 300%, it may also be expressed as “a 3-fold increase”.
A retrospective study relies on data on exposures and/or outcomes that have already been collected (through medical records or as part of another study). Data used in this way may not be as reliable as data collected prospectively as it relies on the accuracy of records made at the time and on people’s recall of events in the past, which can be inaccurate (referred to as recall bias).
A secondary analysis is when researchers revisit data that was collected for a different reason and analyse it again to answer a new research question. This type of analysis is sometimes prone to errors.
Selection bias is a distortion of evidence or data that arises from the way that the data is collected.
This is one of a set of measures used to show the accuracy of a diagnostic test (see specificity, negative predictive value and positive predictive value). Sensitivity is the proportion of people with a disease who are correctly identified as having that disease by the diagnostic test. For example, if a test has a sensitivity of 90%, this means that it correctly identified 90% of the people with the disease, but missed 10% (these people were ‘false negatives’ on the test).
Single nucleotide polymorphism (SNPs)
The human genome is the entire sequence of genetic information contained within our DNA. This sequence is made up of strings of molecules called nucleotides, which are the building blocks of DNA. There are four nucleotides, called A,C, T and G.
All humans share a very high level of similarity in their DNA sequence, particularly within genes, where the sequence of nucleotides contains the instructions for making the proteins that the cell and organism need. However, there are points in the DNA where different people have a different nucleotide, these are called single nucleotide polymorphisms (SNPs, pronounced “snips”).
Most SNPs do not affect a person’s health or characteristics, as they do not lie in parts of DNA that encode proteins. However, they are useful to researchers, as SNPs that are more common in people who have a specific condition than those without the condition indicate that the regions of DNA surrounding these SNPs are likely to contain genes that are contributing to these diseases.
This is one of a set of measures used to assess the accuracy of a diagnostic test (see sensitivity, negative predictive value and positive predictive value). Specificity is the proportion of people without a disease who are correctly identified as not having that disease by the diagnostic test. For example, if a test has a specificity of 95%, this means that it correctly identified 95% of the people who did not have the disease, but that 5% of people without the disease were incorrectly diagnosed as having the disease (these people were ‘false positives’ on the test).
The standard deviation is a statistical term that measures how much individual scores of a given group vary from the average (mean) score of the whole group. Another way of saying this is that it measures the spread of the individual results around the average of all the results.
If the results of a test have statistical significance, it means that they are not likely to have occurred by chance alone. In such cases, we can be more confident that we are observing a ‘true’ result.
This is a synthesis of the medical research on a particular subject. It uses thorough methods to search for and include all or as much as possible of the research on the topic. Only relevant studies, usually of a certain minimum quality, are included.
Time trend studies
Time trend studies are epidemiological studies that describe characteristics of a population over time. They look at trends at the population level (rather than in individuals) through taking repeated cross sectional samples.
Tissue engineering is an interdisciplinary field that applies the principles of engineering and biological sciences to developing functional substitutes for damaged tissue.
Twin studies rely on comparing the phenotypes (observable physical traits) of monozygotic (genetically identical) twins and dizygotic (non-identical) twin pairs. The difference in correlation between phenotypes in the identical twins and the correlation in phenotypes in the non-identical twins estimate the genetic contribution to variations in phenotype (the within-twin correlation).
Water maze test
A water maze test comprises a pool of water, with a single platform (sometimes more than one platform) placed just below the surface of the water. Usually the platform and the pool are white, making the platform difficult to see. Mice are placed in the pool and swim around until they find the platform.
Researchers usually time how long their test mice take to find the platform, but they may also film the mice to examine their searching pattern or technique. This can be an important indicator of their behavioural functions. Usually, mice are tested over and over again to see if they learn where the platform is. If the mice fail to find the platform after a certain time they are usually removed to prevent them from drowning.