Core Revision
Top statistics questions trainees miss
2 June 2026 · reviewed by a Core Revision GP
Statistics is reliably scored once it clicks. Here are the ones people lose marks on — and the trap each time.
← All Core WisdomStatistics is the most reliably scored part of the AKT — once it clicks — because the answers come from understanding what each measure means, not from calculating under pressure. Here are the ones trainees lose marks on most often, and the trap each time. For worked examples of each, our free AKT Statistics Guide goes step by step.
Tests: ruling in and ruling out
- Sensitivity — a highly sensitive test, when negative, rules out (SnNOUT). Specificity — a highly specific test, when positive, rules in (SpPIN).
- PPV and NPV depend on prevalence. The classic trap: the same test has a much lower positive predictive value in a low-prevalence population. If the stem changes the prevalence, the predictive value changes.
- Likelihood ratios shift pre-test to post-test probability; LR+ >10 or LR− <0.1 are clinically useful.
Risk and benefit
- Relative vs absolute risk. When the question asks “by how much?” the answer is almost always the absolute risk reduction. A “50% reduction” can be a fall from 2% to 1% — tiny in absolute terms.
- NNT = 1 / ARR (and NNH = 1 / ARI). Lower NNT = more effective. Be ready to compute it from an absolute risk reduction.
- Odds ratio vs relative risk — close when the outcome is rare, divergent when it’s common.
Precision and study design
- Confidence intervals: a 95% CI that crosses 1 (for a ratio) or 0 (for a difference) is not statistically significant. A wide CI means an imprecise estimate.
- p-values tell you about chance, not effect size or clinical importance.
- Hierarchy: systematic review/RCT > cohort > case-control > case series. Match the design to the question (case-control for rare diseases; cohort for incidence).
- Bias: selection, recall (classic in case-control), lead-time and length-time bias (classic in screening), confounding.
- Intention-to-treat preserves randomisation — analyse people in the group they were assigned to.
Screening is a favourite because it stacks several of these at once: low prevalence (so PPV is low), plus lead-time and length-time bias that make screening look better than it is. If a stem is about a screening programme, those are the marks.
“By how much?” → absolute risk. “How many to treat?” → NNT = 1/ARR. “Does it work in this clinic?” → PPV and prevalence. Read the question for which one it’s actually asking — the wrong measure is always one of the options.
Don’t try to calculate fast — try to interpret correctly. The AKT rarely needs heavy arithmetic; it needs you to pick the right concept and know what it means for the patient in front of you. Learn the one-line meaning of each measure and you’ll bank these marks calmly.
Dr Victoria Hammond · Co-Founder & Lead for Educational Design
Work them as a set, not in isolation — the exam mixes them in a single vignette, and knowing how they interact is what separates a guess from a mark.
Sitting the AKT? Try twenty questions free.
Twenty original AKT questions in the exam's 80/10/10 split: sixteen clinical, two on evidence-based practice and two on organisation and management, with the reasoning after every one and a marked report. Free, no card and the first five are open to all.