Biostatistics in Clinical Research
Expert-defined terms from the Advanced Certificate in Clinical Research course at London School of Planning and Management. Free to read, free to share, paired with a professional course.
Absolute Risk Reduction (ARR) #
Absolute Risk Reduction (ARR)
Definition #
The difference in event rates between a control group and an experimental group, expressed as a proportion.
Example #
If 10 % of patients in the control arm experience a heart attack versus 6 % in the treatment arm, ARR = 0.10 − 0.06 = 0.04 (4 %).
Practical application #
Used to convey the clinical impact of an intervention and to calculate the number needed to treat (NNT = 1/ARR).
Challenges #
Requires accurate event rates; small sample sizes can produce unstable ARR estimates and wide confidence intervals.
Adjusted Hazard Ratio (aHR) #
Adjusted Hazard Ratio (aHR)
Definition #
A hazard ratio derived from a Cox proportional hazards model that includes covariates to control for confounding variables.
Example #
In a cancer trial, the aHR for death after adjusting for age, stage, and performance status might be 0.75, indicating a 25 % reduction in hazard after adjustment.
Practical application #
Provides a more realistic estimate of treatment effect when baseline characteristics differ between groups.
Challenges #
Model misspecification, violation of proportional hazards assumption, and multicollinearity among covariates can bias the aHR.
Analysis of Covariance (ANCOVA) #
Analysis of Covariance (ANCOVA)
Definition #
A statistical technique that combines analysis of variance with linear regression to compare group means while adjusting for continuous covariates.
Example #
Comparing post‑treatment blood pressure between two drug groups while adjusting for baseline blood pressure.
Practical application #
Increases statistical power by reducing residual variance and controlling for baseline imbalances.
Challenges #
Requires linear relationship between covariate and outcome, homogeneity of regression slopes, and careful selection of covariates to avoid over‑adjustment.
Attrition Bias #
Attrition Bias
Definition #
Systematic differences between participants who complete a study and those who withdraw, potentially distorting results.
Example #
If sicker patients are more likely to drop out, the remaining sample may appear healthier than the true population.
Practical application #
Recognized during trial design by planning strategies such as intention‑to‑treat analysis and robust follow‑up procedures.
Challenges #
Quantifying the bias is difficult; high attrition rates (>20 %) often necessitate sensitivity analyses.
Baseline Characteristics #
Baseline Characteristics
Definition #
Demographic and clinical variables measured before randomization, used to assess group comparability.
Example #
Age, sex, disease severity, and prior therapies recorded at enrollment.
Practical application #
Inform stratified randomization schemes and serve as adjustment variables in multivariable models.
Challenges #
Imbalance may occur by chance; over‑adjustment can reduce precision.
Bayesian Inference #
Bayesian Inference
Definition #
A statistical paradigm that updates prior beliefs with observed data to obtain a posterior distribution for parameters of interest.
Example #
Using a prior distribution for treatment effect based on earlier phase II data and combining it with phase III results to produce a posterior estimate.
Practical application #
Facilitates adaptive trial designs, interim monitoring, and decision‑making under uncertainty.
Challenges #
Choice of prior can be subjective; computationally intensive for complex models.
Binomial Distribution #
Binomial Distribution
Definition #
Probability distribution describing the number of successes in a fixed number of independent yes/no trials with constant success probability.
Example #
Number of patients achieving tumor response out of 50 treated individuals.
Practical application #
Basis for confidence interval calculations for proportions and for exact tests (e.g., Fisher’s exact test).
Challenges #
Assumes independence; violations occur with clustered or longitudinal data.
Censoring #
Censoring
Definition #
Incomplete observation of an event time, where the exact time of occurrence is unknown beyond a certain point.
Example #
A patient who is still alive at study end is right‑censored at that time.
Practical application #
Handled using Kaplan‑Meier estimator and Cox models to incorporate all available information.
Challenges #
Informative censoring can bias estimates if the censoring mechanism is related to the outcome.
Confidence Interval (CI) #
Confidence Interval (CI)
Definition #
A range of values constructed from sample data that, with a specified confidence level (typically 95 %), is expected to contain the true population parameter.
Example #
A 95 % CI for a mean difference of 2.5 mg/dL might be (1.0, 4.0).
Practical application #
Provides information about precision and statistical significance; intervals that exclude the null value imply significance.
Challenges #
Misinterpretation as probability that the true value lies within the interval; dependence on sample size and variance.
Cox Proportional Hazards Model #
Cox Proportional Hazards Model
Definition #
A semiparametric regression model that estimates the effect of covariates on the hazard function without specifying the baseline hazard.
Example #
Modeling time to disease progression while adjusting for treatment, age, and biomarker status.
Practical application #
Generates adjusted hazard ratios for multiple predictors in time‑to‑event analyses.
Challenges #
Requires proportional hazards assumption; violation necessitates stratified models or time‑dependent covariates.
Cross‑Over Design #
Cross‑Over Design
Definition #
A clinical trial where each participant receives multiple interventions sequentially, serving as his/her own control.
Example #
Patients receive Drug A for eight weeks, undergo a two‑week washout, then receive Drug B for eight weeks.
Practical application #
Increases efficiency and reduces variability, especially for chronic stable conditions.
Challenges #
Carry‑over effects, appropriate washout duration, and ethical concerns when disease progression is rapid.
Data Monitoring Committee (DMC) #
Data Monitoring Committee (DMC)
Definition #
An independent group of experts tasked with reviewing accumulating trial data for safety, efficacy, and integrity.
Example #
The DMC recommends early termination of a trial because of overwhelming benefit.
Practical application #
Ensures participant protection and objective decision‑making during a study.
Challenges #
Maintaining confidentiality, avoiding operational bias, and defining stopping rules a priori.
Effect Size #
Effect Size
Definition #
A quantitative measure of the magnitude of a treatment effect, independent of sample size.
Example #
A Cohen’s d of 0.8 indicates a large effect of the intervention on depression scores.
Practical application #
Guides sample‑size calculations and facilitates meta‑analysis across studies.
Challenges #
Selection of appropriate metric; effect sizes can be inflated in small, underpowered studies.
Endpoint #
Endpoint
Definition #
The specific event or measurement used to assess the efficacy of an intervention.
Example #
Overall survival, progression‑free survival, or change in HbA1c.
Practical application #
Determines statistical analysis plan and regulatory approval criteria.
Challenges #
Choosing clinically meaningful endpoints versus feasible surrogate markers; endpoint adjudication may be resource‑intensive.
Enrollment #
Enrollment
Definition #
The process of enrolling eligible participants into a clinical trial.
Example #
A multicenter oncology study enrolls 500 patients over 12 months.
Practical application #
Impacts study timelines, power, and budget; strategies include site selection and outreach.
Challenges #
Slow accrual, competition with other trials, and stringent eligibility criteria.
Epidemiologic Measures #
Epidemiologic Measures
Definition #
Quantitative descriptors of disease occurrence in a defined population.
Example #
Incidence rate of 5 cases per 1,000 person‑years for a rare disease.
Practical application #
Provides baseline risk estimates for sample‑size calculations and contextualizes trial results.
Challenges #
Accurate denominator determination and accounting for under‑reporting.
Exponential Distribution #
Exponential Distribution
Definition #
A continuous probability distribution often used to model time between events in a Poisson process, characterized by a constant hazard rate.
Example #
Modeling time to equipment failure in a clinical laboratory.
Practical application #
Serves as a simple parametric alternative to non‑parametric survival methods.
Challenges #
Assumes constant hazard, which is rarely true for disease progression.
Fisher’s Exact Test #
Fisher’s Exact Test
Definition #
A statistical test that calculates the exact probability of observing a particular set of frequencies in a 2 × 2 table, regardless of sample size.
Example #
Comparing adverse event rates (5/30 vs 12/30) between two treatment arms.
Practical application #
Preferred when expected cell counts are <5.
Challenges #
Computationally intensive for larger tables; interpretation identical to chi‑square when sample size is large.
Hazard Ratio (HR) #
Hazard Ratio (HR)
Definition #
The ratio of hazard rates between two groups at any point in time, derived from survival analysis.
Example #
An HR of 0.65 indicates a 35 % reduction in hazard for the treatment group compared with control.
Practical application #
Commonly reported in oncology trials to quantify treatment benefit.
Challenges #
Requires proportional hazards; non‑proportionality leads to misleading single‑value HRs.
Intention‑to‑Treat (ITT) Principle #
Intention‑to‑Treat (ITT) Principle
Definition #
An analysis strategy that includes all randomized participants in the groups to which they were assigned, regardless of adherence.
Example #
A participant who discontinues therapy after two weeks is still counted in the ITT analysis.
Practical application #
Preserves randomization benefits and provides a conservative estimate of treatment effect.
Challenges #
Missing data handling; may dilute true efficacy if non‑adherence is high.
Kaplan‑Meier Estimate #
Kaplan‑Meier Estimate
Definition #
A non‑parametric method for estimating the survival function from time‑to‑event data, accounting for censored observations.
Example #
Plotting the probability of remaining event‑free over 24 months for a new drug.
Practical application #
Visual comparison of survival between groups and basis for log‑rank test.
Challenges #
Does not adjust for covariates; limited to descriptive analysis.
Logistic Regression #
Logistic Regression
Definition #
A regression model that predicts the log‑odds of a binary outcome as a linear function of predictor variables.
Example #
Modeling probability of treatment response based on age, gender, and baseline disease severity.
Practical application #
Generates adjusted odds ratios for risk factor analysis and prediction models.
Challenges #
Requires sufficient events per variable; multicollinearity and separation can impede model convergence.
Mean #
Mean
Definition #
The sum of a set of numeric values divided by the number of observations.
Example #
Mean systolic blood pressure of 128 mmHg in a trial cohort.
Practical application #
Central tendency measure for continuous outcomes; used in t‑tests and ANOVA.
Challenges #
Sensitive to outliers; may not represent skewed distributions.
Median #
Median
Definition #
The middle value separating the higher half from the lower half of a data set.
Example #
Median time to progression of 9 months in a cancer study.
Practical application #
Preferred for skewed data or when outliers are present; basis for non‑parametric tests.
Challenges #
Does not convey distribution shape; less efficient than mean when data are normal.
Mixed‑Effects Model #
Mixed‑Effects Model
Definition #
A statistical model that incorporates both fixed effects (population‑level) and random effects (subject‑specific) to handle correlated or clustered data.
Example #
Analyzing repeated blood pressure measurements across multiple clinics, with random intercepts for each clinic.
Practical application #
Allows inclusion of all available data, accommodates missingness under MAR, and models intra‑subject correlation.
Challenges #
Requires correct specification of random‑effects structure; computationally demanding for large datasets.
Null Hypothesis (H₀) #
Null Hypothesis (H₀)
Definition #
A default statement that there is no effect or difference between groups, against which evidence is evaluated.
Example #
H₀: μ₁ = μ₂ (no difference in mean outcome between treatments).
Practical application #
Forms the basis of p‑value computation; rejection leads to claim of statistical significance.
Challenges #
Misinterpretation as proof of no effect; dependence on sample size.
Odds Ratio (OR) #
Odds Ratio (OR)
Definition #
The ratio of odds of an event occurring in the treatment group to the odds in the control group.
Example #
An OR of 2.0 indicates twice the odds of response with the experimental therapy.
Practical application #
Frequently reported in case‑control studies and logistic regression outputs.
Challenges #
Overestimates risk when outcome is common; interpretation less intuitive than risk ratio.
Paired t‑Test #
Paired t‑Test
Definition #
A statistical test that compares the means of two related groups, accounting for the paired nature of observations.
Example #
Comparing baseline and 12‑week cholesterol levels in the same participants.
Practical application #
Increases power by reducing variability due to subject‑specific factors.
Challenges #
Assumes normality of differences; not appropriate for non‑continuous outcomes.
Power #
Power
Definition #
The probability of correctly rejecting the null hypothesis when a true effect exists; commonly set at 80 % or 90 %.
Example #
A study designed with 90 % power to detect a hazard ratio of 0.75.
Practical application #
Drives sample‑size calculations; higher power reduces risk of false‑negative conclusions.
Challenges #
Over‑estimation of effect size leads to under‑powered studies; increasing power inflates cost and recruitment burden.
P‑value #
P‑value
Definition #
The probability of observing data as extreme as, or more extreme than, those observed, assuming the null hypothesis is true.
Example #
A p‑value of 0.03 indicates a 3 % chance of the observed difference arising by random chance.
Practical application #
Determines whether results cross a pre‑specified significance threshold (e.g., α = 0.05).
Challenges #
Does not measure effect size or clinical relevance; susceptible to misuse and p‑hacking.
Randomization #
Randomization
Definition #
The process of assigning participants to treatment arms using a random mechanism to prevent selection bias.
Example #
A computer‑generated permuted block randomization with block size 4.
Practical application #
Balances known and unknown confounders across groups, supporting causal inference.
Challenges #
Implementation errors, lack of allocation concealment, and potential for imbalance in small trials.
Regression Analysis #
Regression Analysis
Definition #
A set of statistical techniques for modeling the relationship between a dependent variable and one or more independent variables.
Example #
Using multiple linear regression to predict change in weight based on diet, exercise, and baseline BMI.
Practical application #
Adjusts for covariates, predicts outcomes, and estimates effect sizes.
Challenges #
Assumptions of linearity, independence, homoscedasticity, and normality must be checked; over‑fitting is a risk.
Sample Size #
Sample Size
Definition #
The number of participants required to achieve a desired power for detecting a pre‑specified effect, given significance level and variability.
Example #
Calculating that 250 patients per arm are needed to detect a 20 % relative risk reduction with 80 % power.
Practical application #
Informs budgeting, timeline, and feasibility assessments.
Challenges #
Inaccurate assumptions about event rates or variance lead to under‑ or over‑powered studies.
Sensitivity #
Sensitivity
Definition #
The proportion of true positives correctly identified by a diagnostic test.
Example #
A biomarker that detects 90 % of patients with disease X.
Practical application #
Critical for evaluating screening tools and case‑finding algorithms.
Challenges #
Trade‑off with specificity; high sensitivity may increase false‑positive rates.
Specificity #
Specificity
Definition #
The proportion of true negatives correctly identified by a diagnostic test.
Example #
A test that correctly classifies 95 % of disease‑free individuals.
Practical application #
Important for confirming disease absence and reducing unnecessary interventions.
Challenges #
Balancing specificity against sensitivity; context‑dependent clinical relevance.
Survival Analysis #
Survival Analysis
Definition #
A collection of statistical methods for analyzing the time until an event of interest occurs, accommodating censored observations.
Example #
Evaluating median overall survival for a new oncology agent.
Practical application #
Enables estimation of survival curves, hazard ratios, and cumulative incidence.
Challenges #
Assumptions about proportional hazards, handling competing risks, and ensuring adequate follow‑up.
Type I Error (α) #
Type I Error (α)
Definition #
The probability of incorrectly rejecting a true null hypothesis; conventionally set at 0.05.
Example #
Concluding a treatment effect when none exists due to random variation.
Practical application #
Determines the threshold for statistical significance.
Challenges #
Multiple testing inflates overall α; controlling family‑wise error may require adjustments (e.g., Bonferroni).
Type II Error (β) #
Type II Error (β)
Definition #
The probability of failing to reject a false null hypothesis; related to study power (1 − β).
Example #
Missing a genuine benefit of a drug because the sample size is too small.
Practical application #
Guides sample‑size planning to achieve acceptable β (often 0.20).
Challenges #
Under‑powered studies increase risk of Type II errors, potentially leading to erroneous conclusions about efficacy.
Unblinded Study #
Unblinded Study
Definition #
A trial in which participants, investigators, or both are aware of the assigned interventions.
Example #
An open‑label extension where all subjects receive the investigational drug after the double‑blind phase.
Practical application #
May be necessary for pragmatic trials or when blinding is infeasible.
Challenges #
Susceptible to performance and detection bias; outcomes may be influenced by knowledge of treatment allocation.
Variance #
Variance
Definition #
A measure of the spread of data points around the mean, calculated as the average squared deviation.
Example #
Variance of systolic blood pressure measurements equal to 225 mmHg².
Practical application #
Essential for sample‑size calculations and for assessing model fit.
Challenges #
Sensitive to outliers; interpretation less intuitive than standard deviation.
Weighted Least Squares (WLS) #
Weighted Least Squares (WLS)
Definition #
A regression technique that assigns weights to observations inversely proportional to their variance, improving efficiency when error variance is unequal.
Example #
Analyzing survey data where larger hospitals contribute more precise estimates than smaller ones.
Practical application #
Corrects for heteroscedasticity and yields unbiased parameter estimates.
Challenges #
Requires accurate variance estimates; misspecified weights can worsen bias.
Yield #
Yield
Definition #
The proportion of screened candidates who become enrolled participants.
Example #
A 30 % yield when 150 out of 500 screened patients consent to join the study.
Practical application #
Assists in forecasting recruitment timelines and budgeting.
Challenges #
Low yield may indicate overly restrictive eligibility or inadequate outreach.