Glossary
Plain definitions of the research terms used on this site. Each term links to the pages and notes where it matters.
A
- Adherence
How consistently people use a tool or follow a study protocol over time, often reported as the share of planned days or sessions completed.
- AUC (area under the ROC curve)
One number from 0.5 (no better than chance) to 1 (perfect) that summarizes how well a score separates two groups.
Related: Calibration
B
- Bias adjustment (brain age)
A correction for the tendency of brain-age models to overestimate age in younger people and underestimate it in older people.
Related: Brain-PAD
- Biomarker
A measurable characteristic, such as a regional brain volume, that indicates a biological process, a disease state or a response to treatment.
- Bland–Altman plot
A chart of the differences between two measurement methods against their average. It shows systematic bias and the limits within which most differences fall.
Related: ICC (intraclass correlation)
- Brain age
An estimate of a person's age made by a statistical model from a brain scan.
Related: Brain-PAD
- Brain-PAD
Brain predicted age difference: brain age minus actual age. A positive value means the brain looks older than expected. It is a research measure, not a diagnosis.
Related: Bias adjustment (brain age)
C
- Calibration
How closely a model's predicted values or risks match what is actually observed.
Related: AUC (area under the ROC curve)
- Cross-validation
Testing a model on parts of the data it was not trained on, by splitting the data repeatedly. Nested cross-validation also keeps model tuning separate from testing.
Related: External validation
E
- Effect size
How large a difference or association is, expressed in a way that does not depend on the sample size.
Related: Statistical power
- Endpoint
The outcome a study or trial is designed to measure, defined before the study starts.
- Enrichment
Selecting trial participants who are more likely to show the outcome, for example by a biomarker, so that a smaller trial can detect an effect.
Related: Biomarker
- Equivalence testing
A test that shows two methods differ by less than a margin set in advance, instead of only failing to find a difference. TOST is the usual method.
Related: Tolerance interval
- External validation
Testing a model on data from a different source, site or period than the data used to build it.
Related: Cross-validation
I
- ICC (intraclass correlation)
A reliability measure from 0 to 1 that shows how consistent repeated measurements are, for example the same test taken twice.
Related: Test–retest reliability
M
- Mild cognitive impairment (MCI)
Cognitive decline greater than expected for a person's age, but not severe enough to interfere much with daily life.
- Missing data
Values that were planned but not collected. How they are handled, for example by multiple imputation, can change the results.
- Mixed-effects model
A regression model for repeated or grouped measurements, such as several visits per person or several participants per site.
- Morphometry
Measuring the size and shape of brain structures from images, such as cortical thickness or regional volume.
Related: T1-weighted MRI
N
- Normative data
Results from a reference group that show which scores are typical for people of a given age, sex or education.
P
- Practice effect
Better scores on a repeated test caused by having taken it before, not by a real change in ability.
Related: Test–retest reliability
R
- Real-world data
Data collected during routine care or daily life rather than in a controlled study.
S
- Sensitivity and specificity
The share of people with a condition whom a test correctly flags (sensitivity), and the share without it whom the test correctly clears (specificity).
Related: AUC (area under the ROC curve)
- Software as a medical device (SaMD)
Software intended for a medical purpose, such as diagnosis or monitoring, which Health Canada and other regulators may regulate as a device.
- Statistical analysis plan (SAP)
A document written before the analysis that fixes the outcomes, the methods and the handling of missing data.
- Statistical power
The probability that a study detects an effect of a given size if the effect is real. It drives the sample size.
Related: Effect size
T
- T1-weighted MRI
A common structural MRI scan with good contrast between grey matter, white matter and fluid; the usual input for brain-age models.
Related: Brain age
- Test–retest reliability
How similar the results are when the same test is repeated under the same conditions.
Related: ICC (intraclass correlation)
- Tolerance interval
A range expected to contain a stated share of all future measurements (for example 95%) with stated confidence. Used to set device acceptance criteria.
Related: Equivalence testing
V
- Verification and validation (V&V)
In device development, verification checks that the device meets its specification; validation checks that it meets users' needs.
Related: Tolerance interval
A clear question is a good place to start.
Start with a short, non-confidential description. We reply with a suggested next step.