Statistical Analysis Terms

Odds ratio

compares the odds of an outcome occurring in one group to the odds in another group.

inoculum effect

Higher bacterial loads can make pathogens less susceptible to certain antibiotics, even when they are normally effective at lower bacterial concentrations

clinical nuance

refers to the subtle, context-dependent judgment that clinicians apply when making decisions about patient care. It goes beyond guidelines or textbook answers by considering the unique details of a patient's situation

surrogate marker

a substitute measure used in clinical trials when the true clinical outcome (such as survival, heart attack, or stroke) takes too long, is too difficult, or is too expensive to measure directly

- Ex: BP, blood cultures

false

T/F: You can extrapolate superiority from a noninferiority study.

noninferioirity

drug A is no worse than drug B

Cost-utility analysis

type of economic evaluation used in healthcare to compare the costs and outcomes of different interventions

Intention-to-Treat (ITT) Analysis

An analytic principle in clinical trials where all participants are analyzed in the groups to which they were originally randomized, regardless of whether they actually completed, adhered to, or even received the assigned treatment

intent to treat

The only method that preserves randomization.

- prevents bias

Power

The probability that a study will correctly detect a true effect (or difference) if one really exists.

80%

good power

90%

great power

Confidence interval (CI)

The range of values denoted by the upper and lower limits that describe the plausible values of a calculated point estimate; this relates to the amount of uncertainty or precision in an estimate of the parameter.

CONSORT (CONsolidated Standards Of Reporting Trials)

An evidence-based set of guidelines designed to improve how randomized controlled trials (RCTs) are reported.

pivotal trial

A large, well-controlled clinical study (usually Phase 3) that provides the key evidence needed to determine whether a drug should be approved by regulatory authorities

arm

a term to describe one of the groups that participants are assigned to, each receiving a specific treatment or control.

validity

the degree to which a measurement truly assess what it is intended ot measure

internal validity

the study is designed and conducted in an appropriate manner so as to arrive at accurate results

external validity

the results of the study are able to be accurately generalized to the larger population

p-value

measures how likely your observed results (or more extreme ones) would be if the null hypothesis were true

- It means:

> Assume the null hypothesis is true (no difference).

> Then ask: “How surprising are my results?”

> The p-value is that probability of being surprised.

central tendency

Used to provide information about the center or "typical value" of a set of numbers.

- "Where do most of the values cluster?"

- Instead of listing every data point, we summarize the dataset with one representative number.

mean, median, mode

Measures of central tendency

median

The _____ is most appropriately used in situations where a few values of a variable are noticeably larger or smaller than most of the rest of the values. These values are known as outliers. In these situations, this measure of central tendency is preferred because it is a more accurate representation of the majority of values present in the data.

prevalence

measures the probability of having a disease at a point in time

- reflects existing disease within a population

incidence

describes the risk of developing a new disease, symptom, or problem.

Morbidity

Incidence and prevalence are measures of ___ (mortality or morbidity).

Interval data

Data where the numbers of a scale represent a rank order and equal differences between numbers represent equal differences on the variable being measured; however, a defined and meaningful zero point is lacking.

- Equal Intervals, No True Zero

- Type of continuous data

- Examples:

> Temperature in Celsius or Fahrenheit (0°C ≠ "no temperature")

> Calendar years (difference between 2000 and 2010 is the same as 2010 and 2020, but year "0" isn't an absence of time)

Ratio data

(ratiO has a zerO)

Data measured on a numeric scale that can be ordered, with equal intervals between values, and there is an absolute minimum or zero point to the scale; there is a defined and meaningful zero point that denotes "none of" the property being measured.

- Equal Intervals + True Zero

- Type of continuous data

- Examples:

> Weight (0 kg = no weight)

> Height (0 cm = no height)

> Age (0 years = newborn)

> Blood pressure, heart rate, cholesterol levels

Nominal data

Data that can be placed into narrowly defined categories that are not in any particular order.

- Names / Categories

- Examples:

> Blood type (A, B, AB, O)

> Gender (male, female)

> Yes/No responses

Ordinal data

Data that consist of narrowly defined categories, but these categories have a rank order; the difference between the ordered categories cannot be considered to be equal.

- Order, but Unequal Intervals

- Examples:

> Pain scale (mild, moderate, severe → or 1-5 Likert scale)

> Cancer stages (I, II, III, IV)

> Class rank (1st, 2nd, 3rd)

Continuous data

Data that can take any value on a scale, often measured rather than counted.

- logical order with values that increase or decrease by the same amount

- Examples: Height, weight, blood pressure, cholesterol, temperature

Student's t-test (independent samples t-test)

A parametric test comparing the means of two independent groups when data are continuous and normally distributed

- Example: comparing average cholesterol in men and women

Mann-Whitney U test (Wilcoxon Rank-Sum Test)

- Think "Mann and Whitney = two independent people."

Nonparametric alternative to the independent t-test; compares medians/rank distributions of two independent groups.

- Example: Pain scores (1-10) for drug vs. placebo when data are skewed

Paired t-test

A parametric test comparing the means of two related groups (same people measured twice or matched pairs)

- Example: Weight before vs. after a diet program.

Wilcoxon Signed-Rank Test

Nonparametric alternative to the paired t-test; compares two related samples when data are not normal

- Example: Quality of life score before vs. after surgery.

ANOVA (Analysis of Variance)

- One-way: independent

- Repeated measures: related groups

Test for statistical significance when using parametric, continuous data with 3 or more samples or groups

- can measure both independent and related groups.

- Example: Compare average cholesterol in patients on Diet A, Diet B, Diet C (different people in each diet group).

- Example: The same patients measured at baseline, 1 month, and 3 months

Kruskal-Wallis Test

Nonparametric alternative to one-way ANOVA; compares medians of three or more independent groups

- Example: Median satisfaction scores across 3 hospitals

Friedman Test

Nonparametric alternative to repeated-measures ANOVA; compares three or more related groups.

- Example: Blood pressure measured at baseline, 1 month, and 3 months in the same patients.

Chi-square Test of Homogeneity

A statistical method for testing the differences between the proportions of two or more groups, where the groups are said to be independent; thus, it involves comparing the groups on a dichotomous outcome variable.

- Compares categorical data — observed vs. expected frequencies.

- “Do different groups have different percentages of people with a yes/no outcome?”

- The test compares the percentages of “yes” outcomes between groups to see if they’re truly different, or if the difference could just be random.

- Example: Does smoking status (yes/no) differ by lung cancer status (present/absent)

Fisher's Exact Test

Exact test for small sample sizes when expected cell counts <5; alternative to Chi-square.

- Example: Comparing rare disease occurrence in treatment vs. placebo group.