Biostatistics, Study Design, and Clinical Trial Analysis

Vocabulary practice flashcards covering study design hierarchies, clinical trial appraisal, biostatistical test selection rules, mathematical formulas, and findings from the JUPITER clinical trial.

Randomized Controlled Clinical Trial (RCT)

The gold standard study design (Hierarchy Rank 1) characterized by the random allocation of participants to experimental or control groups, which minimizes both known and unknown confounding variables to establish causal inference.

Non-randomized Controlled Clinical Trial

A quasi-experimental study design (Hierarchy Rank 2) that utilizes a control group but lacks random allocation, leaving it vulnerable to selection bias while evaluating risk using Relative Risk ($\text{RR}$) or Hazard Ratio ($\text{HR}$).

Prospective Cohort Study

An observational longitudinal study design (Hierarchy Rank 3) that follows exposed and unexposed groups forward in time to establish temporal sequence, utilizing Relative Risk ($\text{RR}$) and Odds Ratio ($\text{OR}$).

Retrospective Cohort Study

An observational longitudinal study design (Hierarchy Rank 4) that identifies cohorts from historical records and traces outcomes forward from that past reference point.

Case-Control Study

An observational retrospective study design (Hierarchy Rank 5) that compares individuals with an outcome (cases) to individuals without the outcome (controls) backward in time; it is optimal for rare diseases and can only measure the Odds Ratio ($\text{OR}$).

Cross-Sectional Study

An observational snapshot design (Hierarchy Rank 6) assessing exposure and outcome at a single point in time; it cannot establish temporality or causality and uses only Odds Ratio ($\text{OR}$).

Case Series Study

A descriptive study design (Hierarchy Rank 7) tracking a small group of patients with similar clinical presentations or treatments without a comparison or control group.

Case Study / Case Report

The lowest-ranked descriptive study design (Hierarchy Rank 8) consisting of an intensive descriptive report on a single patient, which is vulnerable to anecdotal interpretation.

Treatment Allocation Concealment

A trial method preventing selection and allocation bias by shielding upcoming assignments from researchers and participants using sequentially numbered opaque sealed envelopes, coded containers, or central computer-generated randomization.

Intention-to-Treat (ITT) Analysis

An analytical approach where all randomized subjects are evaluated in the groups to which they were originally assigned regardless of non-compliance, protocol deviations, or withdrawal, thereby preserving randomization and reflecting real-world community efficacy.

Differential Dropout Rate (Fatal Flaw Threshold)

A drop-out imbalance of $\ge 15\%$ between study arms that introduces severe attrition/selection bias resulting in a "Poor" study quality rating (acceptable trial thresholds require overall dropout $\le 30\%$ and differential dropout $< 15\%$).

Null Hypothesis ($H_0$)

The foundational assumption in statistical testing that specifies no effect, no difference, or no association between variables in the underlying population.

Two-Sample Independent t-Test

A parametric test used to compare the means of a continuous outcome variable between two independent categorical groups (e.g., Male vs. Female).

One-Way Analysis of Variance (ANOVA)

An omnibus hypothesis test used to compare the means of a single continuous outcome variable across three or more independent categorical groups (e.g., Single vs. Married vs. Divorced).

Type I Error Inflation

The compounding increase in the cumulative probability of falsely rejecting a true null hypothesis when conducting multiple simultaneous pairwise tests instead of a single omnibus test like ANOVA.

Bonferroni Correction

A post-hoc adjustment formula, calculated as $\text{Adjusted }\alpha = \frac{\alpha_{\text{original}}}{\text{number of tests}}$, used to control family-wise Type I error inflation during multiple simultaneous comparisons.

F-Statistic

The omnibus test statistic in ANOVA calculated as $F = \frac{\text{Variance Between Groups}}{\text{Variance Within Groups}}$.

Chi-Square Test ($\chi^2$)

A non-parametric test used when both the predictor and outcome variables are categorical, comparing observed frequencies against expected frequencies in a contingency table via $\chi^2 = \sum \left[ \frac{(O - E)^2}{E} \right]$.

Expected Cell Frequency ($E$)

The theoretical frequency of an individual contingency table cell calculated under the null hypothesis as $E = \frac{\text{Row Total} \times \text{Col Total}}{\text{Grand Total}}$.

Paired t-Test

A statistical test that compares the means of a continuous outcome measured in the same group of subjects across two different time points (e.g., Baseline vs. End).

Pearson Correlation Coefficient ($r$)

A measure of the linear direction and strength of association between two continuous variables, ranging strictly from $-1.0$ (perfect negative association) to $+1.0$ (perfect positive association).

Coefficient of Determination ($R^2$)

The squared value of the correlation coefficient ($r^2$), representing the proportion of total variance in the continuous dependent variable ($Y$) explained by the predictor variable(s) ($X$), where $R^2 \ge 0.70$ typically indicates a strong explanatory model.

Simple Linear Regression (SLR)

A statistical model ($Y = \beta_0 + \beta_1 X_1 + \epsilon$) that assesses how a single predictor variable ($X$) predicts a continuous outcome variable ($Y$), testing the null hypothesis $H_0: \beta_1 = 0$.

Multiple Linear Regression (MLR)

A statistical model ($Y = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \dots + \beta_n X_n + \epsilon$) used to estimate the direct, adjusted effect of each predictor on a continuous outcome variable while holding all other predictors constant, testing $H_0: \beta_1 = \beta_2 = \dots = \beta_n = 0$.

Logistic Regression

A regression modeling technique used to predict a binary categorical outcome variable ($0/1$ or Yes/No) from one (simple) or multiple (multiple) predictor factors.

Kaplan-Meier Test

A non-parametric time-to-event method used to calculate and compare cumulative survival distributions across two or more independent groups.

Cox Proportional Hazard Regression

A survival analysis regression model that evaluates the effect of one or multiple predictor variables on time-to-event outcomes, yielding hazard ratios without requiring knowledge of the baseline hazard shape.

Relative Risk (RR)

The ratio of outcome incidence in the exposed group to outcome incidence in the unexposed group, calculated in prospective cohorts and clinical trials as $\text{RR} = \frac{a / (a + b)}{c / (c + d)}$.

Odds Ratio (OR)

A measure of association calculated from a $2 \times 2$ table as $\text{OR} = \frac{a \cdot d}{b \cdot c}$, representing the ratio of the odds of exposure among cases to the odds of exposure among controls.

Confidence Interval Null Value (Relative Risk / Odds Ratio)

The value of $1.0$ for ratio measures; if a $95\%$ confidence interval includes $1.0$, the result is not statistically significant, whereas excluding $1.0$ indicates statistical significance.

Pooled Variance ($s_p^2$)

The weighted average variance across two groups in an independent t-test with equal variances assumed, computed as $s_p^2 = \frac{(n_1 - 1)s_1^2 + (n_2 - 1)s_2^2}{n_1 + n_2 - 2}$.

JUPITER Trial

The "Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin," a randomized, double-blind, placebo-controlled trial of $17,802$ individuals without hyperlipidemia (LDL $< 130\,\text{mg/dl}$) but with elevated hs-CRP ($\ge 2.0\,\text{mg/liter}$) assigned to rosuvastatin ($20\,\text{mg}$ daily) or placebo.

High-Sensitivity C-Reactive Protein (hs-CRP)

An inflammatory biomarker that independently predicts future vascular events; in the JUPITER trial, baseline levels $\ge 2.0\,\text{mg/liter}$ defined study eligibility, and treatment with rosuvastatin lowered median levels by $37\%$.

JUPITER Primary End Point Hazard Ratio

The relative risk of the composite primary cardiovascular end point with rosuvastatin compared to placebo, observed as $\text{HR} = 0.56$ ($95\%$ CI, $0.46\text{--}0.69$; $P < 0.00001$), representing a $44\%$ relative reduction in events.

Number Needed to Treat (NNT in JUPITER)

The number of patients who must be treated with rosuvastatin to prevent one primary composite end point, determined to be $95$ patients for $2$ years, $31$ patients for $4$ years, and $25$ patients projected over an average $5$-year period.

Physician-Reported Diabetes (JUPITER Trial)

A documented adverse event that occurred significantly more often in the rosuvastatin arm ($270$ cases, $3.0\%$) than the placebo arm ($216$ cases, $2.4\%$; $P = 0.01$), accompanied by a modest increase in median glycated hemoglobin ($5.9\%$ vs. $5.8\%$; $P = 0.001$).