#separator:Tab
#html:false
#tags column:5
IMF-101-03-0001-en	Why does adjusting a regression coefficient for other variables not make it causal?	Adding variables does not automatically remove bias; what each parameter means depends on the outcome scale, functional form, sampling process and variable timing.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0002-en	Why is categorising a continuous predictor usually inferior to splines, polynomials or transformations?	Categorising loses information, reduces power and creates threshold interpretations dependent on chosen cut points, whereas the alternatives represent curvature without arbitrary categories.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0003-en	Why can a model show no interaction on one scale yet interaction on another?	Interaction changes an additive effect in a linear model but the effect on that model's own scale in logistic or survival models, so absence on one scale does not imply absence on another.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0004-en	Why can logistic regression output exaggerate an effect, and what is often more interpretable?	Exponentiated coefficients are odds ratios, not risk ratios, and with common outcomes they exaggerate relative change; predicted probabilities are clearer but depend on intercept and covariate pattern.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0005-en	How can adjustment itself introduce bias into an estimated effect?	Adjusting for a consequence of exposure removes part of the effect, and adjusting for a common effect of two variables creates collider bias.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0006-en	Why can simple survival estimates be biased when sicker participants leave follow-up differently?	Standard methods usually assume censoring is conditionally independent of the event process, and differential loss of sicker participants violates that assumption.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0007-en	What does the Cox model assume, and what can communicate effects better when hazards vary or cross?	It leaves the baseline hazard's shape unspecified but assumes a hazard ratio constant over time; restricted mean survival time or time-specific absolute risks may then be clearer.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0008-en	How does treating competing deaths as censored differ in what it estimates from cumulative-incidence methods?	Censoring competing deaths estimates a hypothetical world where that event is removed, under strong assumptions; cumulative incidence estimates event probability in the observed world with competition.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0009-en	Why can a prediction model with good discrimination still be unsuitable for use?	Discrimination only shows that higher-risk people tend to have more events; calibration, agreement of predicted and observed risk, can still fail by systematic overprediction or in important subgroups.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0010-en	Why does choosing a threshold by maximising a generic index risk mismatching clinical practice?	It assumes false positives and false negatives have particular equal costs, and the ROC curve behind it ignores prevalence and the consequences of errors.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0011-en	Why must cross-validation or bootstrapping repeat the entire modelling process to estimate optimism?	A model learns sample-specific noise, so training performance is optimistic; optimism is estimated only when preprocessing and selection are also repeated within each resample.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0012-en	Why can a locked model perform differently on external validation even though it is computed identically?	External validation tests transportability: case mix, measurement, treatment, prevalence or biological relationships can differ in new participants, settings or times.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0013-en	Why can a statistical test not establish whether data are missing at random or not at random?	These mechanisms are assumptions about a process, and missing not at random depends on unseen values; reasons, timing and patterns must be investigated through design knowledge.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending
IMF-101-03-0014-en	Why does multiple imputation represent uncertainty more honestly than single imputation?	Single imputation treats filled values as known and understates uncertainty; multiple imputation analyses several plausible completed datasets and combines within- and between-imputation uncertainty.	Integrated Medical Foundations; chapter 101, TTS module 3; https://theisegoria.github.io/medical-textbook/integrated-medical-foundations-expanded-english.md; module SHA256 at export: 9473452c9e296d3b0a9627fdb6a18492d9de58ba14730207a49faedb22396cdd; editorial review pending	imf lang::en chapter::101 module::101-03 review::pending