Absolute Risk Reduction — The risk associated with one treatment group directly compared (minus) to the risk in another treatment group. The absolute value of the difference in risk between two treatment groups.
Alpha / Type I Error — Pre-determined before the study. The probability of concluding that there is a difference among groups when there may not be.
Attributable Risk — Risk of disease or outcome in those exposed minus the risk in the unexposed.
Beta / Type II Error — The probability of concluding that no difference exists among groups when there may be a difference.
Bias — An unintended consequence that influences the intervention or outcome. Some types of bias are systematic and include the following:
- Publication Bias — The publication of literature has a tendency to support positive results. An important issue in meta-analysis.
- Recall Bias — Study subjects may not recall historical events that are required in surveys.
- Selection Bias — The scheme for randomizing subjects to the treatment arms (intervention groups) may systematically favor/disfavor individuals with certain characteristics.
Blinding— Blinding is used to reduce bias in clinical studies. When both the investigator and study subjects are unaware of the intervention being received it is called double-blinding.
Case-Control — Studies where identifying the impact of an exposure or intervention are of interest. An investigator creates identifies groupings of subjects based on a disease or outcome and then investigates what exposures or interventions led to the outcome.
Clinical Significance — Differs from statistical significance. Clinical significance takes real-world exposure when determining whether a treatment could be effective in a patient population. Although statistical significance may be achieved the clinical benefit of an intervention may not outweigh the risks or increase the quality of life.
Cohort — A term used to describe the grouping of subjects in a study. Can be interchanged with “treatment arm”. These studies can be prospective or retrospective if historical data is used to split groups of subjects into those that received some exposure to identify if an outcome of interest occurred.
Confidence Interval — An interval estimate of a population parameter. Investigators usually state that they are 1-alpha (e.g., 95%) confident that repeated sampling will produce such intervals that include the true parameter value (e.g., mean). Narrow intervals are more precise than larger intervals. However, there is no probability associated with the true mean being within the interval.
Confounder — A factor that affects both the dependent and independent variables. The factor is related to both the intervention and outcome causing often unknown and unintended effects on the outcome.
Cross Over Study — Studies in which subjects serve as their own control. Each subject receives both treatments. There should be a period between treatments that serves to “wash-out” or reduce the effects of each treatment.
Cross Sectional Study — A study conducted to understand a population at a specific time. A snapshot.
Dependent Variable — Outcome variable; Y.
Effectiveness — Measure of practical use in actual or real-world cases of a particular intervention.
Efficacy — Measure of impact of a intervention restricted by ideal conditions such as those of a clinical trial or study.
Hazard Ratio — A ratio of the rates of hazards among two populations.
Heterogeneity — An assessment of the variability among estimates. As compared to homogeneity where estimates are relatively not variable.
Hypothesis Test Statistic — Statistic used to test the validity of a theory. Used to determine the degree of belief of the contention in question.
Incidence (of Disease or Conditions) — The probability or rate or number of new cases of a condition with respect to a period of time.
Independent Variable — Risk factor; exposure; or explanatory variable; X.
Intent to Treat — In a clinical study attempts to increase the generalizability because the study subjects are included in the analysis regardless of study completion.
Meta-Analysis — A method of pooling data and estimates from several research studies in order to increase generalizability of outcomes of interest.
Negative Predictive Value — True negative / (true negatives + false negatives)
Null Hypothesis — The contention or hypothesis that there is no difference among measures of interest.
Number Needed to Harm — This is the number of treated patients needed for an ill outcome of interest. Is equal to 1 / (attributable risk difference).
Number Needed to Treat — Also the number of treated patients needed for an outcome of interest. Is equal to 1 / (absolute risk difference).
Odds Ratio — The odds of disease given exposure / the odds of disease given unexposed. Usually used for rare disease to compare the effect of an exposure.
Per Protocol — In a clinical study the subgroup of patients analyzed that completed the study and adhered to treatment protocols. Attempts to understand the treatment effect at the expense of external validity.
Period Prevalence — The proportion of a population that has a condition over a set period of time
Point Estimate — An estimate of the true population parameter using a sample.
Point Prevalence — The proportion of a population that has a condition at a certain point in time.
Positive Predictive Value — True positives / (true positives + false positives).
Power — The likelihood that a trial will detect a statistical difference. Equivalent to one minus the probability of type II error.
P-Value — The probability that a model parameter is more extreme than what was estimated from a sample. Provides no information about the strength of effect or association, the probability of the null hypothesis being true, nor the chance of making incorrect assumptions.
Receiver Operating Characteristics — In diagnostic testing. A graph that depicts the true positive rate against the false positive rate.
Relative Risk — Risk reduction describes the expected decrease of the rate when comparing two treatment arms in a clinical study. The probability of an event in the treatment group compared (divided by) to the probability in the control arm. If the relative risk is close to 1 then the intervention may not have a great impact.
Relative Risk Reduction — The reduction of risk in a particular treatment arm compared to a baseline risk of a control group. Calculated as (1 - relative risk).
Sample Size — Determined before beginning a study. A representative cohort of the population of study.
Sensitivity — The true positive rate. True positives / (true positives + false negatives)
Specificity — The true negative rate. True negatives / (true negatives + false positives)
Subgroup Analysis — Subdivision of the population based on subject characteristics such as age, sex, or condition.
Surrogate Endpoint — A measure or endpoint that is correlated with the true endpoint and is also assumed to have a strong relationship. Surrogate endpoints may be used in clinical trials instead of true endpoints for efficiency and feasibility reasons.
Time-to-Event — Also known as survival analysis. Attempts to analyze and understand the distribution of times until an event occurs.
Type II Error — The probability of deciding to not reject the null hypothesis when the alternate may be true. Is equal to 1 - power.
Wald Test — A statistical test used for the difference among measurements.