{"entity":{"id":"denominators-and-conditional-probability","kind":"term","name":"Denominators and conditional probability","aka":["conditional probability","denominator"],"tldr":"A percentage answers a question about a particular group. Name that group before interpreting the number, and keep unknown outcomes separate from measured non-response.","summary":"A proportion identifies an outcome within a defined group. The denominator might be everyone enrolled, everyone treated, people with an evaluable scan, or only people with a recurrence; those answer different questions. Conditional probability makes the specified group explicit and is not a statement of causation. Repeated measurements do not create new independent people. Unknown outcomes and incomplete follow-up require explicit handling rather than being silently treated as measured non-response.","asOf":"2026-11-10","links":[{"label":"Hernan and Robins: Causal Inference: What If","url":"https://miguelhernan.org/whatifbook"},{"label":"CONSORT 2025: explanation and elaboration","url":"https://www.bmj.com/content/389/bmj-2024-081124"},{"label":"FDA: oncology clinical-trial endpoints (2018)","url":"https://www.fda.gov/media/71195/download"}],"tags":[],"related":["confidence-interval","hazard-ratio","kaplan-meier-curve"],"cancers":[],"sections":["drug-discovery"],"technologies":[],"targets":[],"drugs":[],"companies":[],"institutions":[],"pathways":[],"terms":[],"trials":[],"people":[],"bottlenecks":[],"keyPapers":[],"journals":[],"dependsOn":[],"notes":["“Half had a response” is incomplete until the counted group is named. The numerator counts observations meeting the event rule; the denominator identifies the relevant group or observation amount. Conditional probability P(A | B) means the probability of A among those satisfying B, with P(A | B) = P(A and B)/P(B) when P(B) is positive. It is undefined if that conditioning probability is zero. “Given” does not mean “caused by”, and the probability of a positive test among people with disease differs from the probability of disease among people with a positive test. People, lesions, blood draws and wells supply different denominators. Events per person-time are a rate, not a simple probability; repeated samples do not become independent patients. A group defined by treatment response may also be selected by underlying biology, so a conditional frequency does not show that the treatment caused the difference. A changed proportion needs its reference proportion attached, because percentages and percentage points answer different questions.","A fictional study enrols twenty people, six never receive the product, fourteen are treated, ten have evaluable scans and five meet the response rule. The observed response proportions are 5/20 = 25% among enrolled people, 5/14, about 35.7%, among treated people, and 5/10 = 50% among those with evaluable scans. The scan-evaluable group is nested within the treated group, which is nested within enrolment. Five other evaluable scans did not meet the response rule; four treated people lack evaluable scans and six were untreated, with no response outcomes supplied for those groups. Calling them confirmed nonresponders would add an assumption. A protocol may classify missing assessments as nonresponse, but that rule and its sensitivity analyses should be reported explicitly.","In a separate complete fictional cohort of 100 people, twenty recur and five of those have brain involvement. Brain involvement is 5/20 = 25% among those with recurrence and 5/100 = 5% in the original cohort. Neither is a personal forecast. In another example with forty enrolled, thirty treated, twenty evaluable and eight responses, 8/20 = 40% describes the evaluable group, whereas 8/40 = 20% describes observed responders among everyone enrolled without establishing the other outcomes. Incomplete follow-up also means recorded events divided by enrolment may not estimate risk at a chosen time; survival or competing-risk methods address defined questions under their assumptions. Safety and efficacy tables can use different analysis sets, and biomarker subgroups or later landmarks change who contributes. Multiple reports from the same participants add no independent people. Exclusion because progression prevented assessment can itself affect interpretation."],"provenance":{"editedBy":"OnCo education audit","editedOn":"2026-11-10"},"category":"Methods and models"},"route":"/terms/denominators-and-conditional-probability/","neighbours":{"term":[{"id":"absolute-benefit","kind":"term","name":"Absolute versus relative benefit (number needed to treat)","route":"/terms/absolute-benefit/"},{"id":"confidence-interval","kind":"term","name":"Confidence interval","route":"/terms/confidence-interval/"},{"id":"hazard-ratio","kind":"term","name":"Hazard ratio (HR)","route":"/terms/hazard-ratio/"},{"id":"kaplan-meier-curve","kind":"term","name":"Kaplan-Meier curve, censoring and proportional hazards","route":"/terms/kaplan-meier-curve/"}],"section":[{"id":"drug-discovery","kind":"section","name":"Drug Discovery Platforms","route":"/fronts/drug-discovery/"}]}}