| B-ID | Part | Title | Guiding question | W-ID | Path status | R lab |
|---|---|---|---|---|---|---|
| I | Clinical decisions and the translational chain | Who will do something differently because of the output? | Medical students, clinical researchers and doctoral candidates new to data analysis | Required (W1 core reading) | Yes: first R steps, pathway data frame, first ggplot2 graphic with expected output, decision-map completeness check, rows vs independent units of both teaching datasets, timing counter-check (4 stopifnot) | fig-b01-pathway (pathway timeline), tbl-b01-map (decision map) |
| I | Describe, predict, explain | Which target quantity answers the actual question? | Clinicians and researchers planning analyses of routine or registry data | Required (W5 core reading) | Yes: dagitty adjustment sets, simulation of 4,000 patients, crude vs adjusted vs ICU-only risk differences, explanation of the sign flip, 100 repetitions, prognostic logistic model, counter-check without the delay-to-ICU arrow (6 stopifnot) | fig-b02-dag (DAG), tbl-b02-questions (three questions) |
| I | How data arise: missingness and denominators | Which patient, animal or sample is missing from the table? | Students and clinical researchers who collect or receive clinical or experimental data | Required (W1 core reading) | Yes: patient and visit tables, dictionary, counts and means per window, two-path count of patients, measurements and linked records, spaghetti plot, denominator figure, CSV export, HSI pseudoreplication, k-anonymity of a registry extract, counter-check with hidden values (7 stopifnot) | fig-b03-trajectories, fig-b03-denominators, tbl-b03-status, tbl-b03-dictionary |
| I | Reconstructing a research project in R | Can a colleague regenerate the number in the figure? | Anyone handing over, reviewing or reusing an analysis, R beginners | Supplementary (recommended reading) | Yes: R objects, relative path, base graphic, long and wide, five scripts executed in memory, two-path counts, target value V4 = 0.56 m/s (n = 7 of 12), SHA-256 comparison with the B03 file, sessionInfo, Quarto report and Git listings, artefact register, firmware counter-check (9 stopifnot) | fig-b04-windows (base R), tbl-b04-artefacts |
| II | From baseline to learning model | Which additional effort delivers an improvement that matters for a decision? | Clinicians, doctoral researchers and reviewers building or appraising a first prediction model | Supplementary (recommended reading for W1 and W3) | Yes: synthetic delirium cohort (1,200 patients, constant IDs, 800/400 split), exported to data/b05-delirium-cohort.csv, 8 stopifnot checks | Fig. 5.1 test AUC with bootstrap intervals and cases found at 30% capacity (n = 400 test patients, 103 events), Table 5.1 where each step happens |
| II | Accuracy, calibration and clinical utility | When does a good ranking lead to the wrong action? | Clinicians, nurses, doctoral researchers and reviewers who set thresholds | Required (W3 core reading) | Yes: reads the B05 test patients (checksum of test IDs), 9 stopifnot checks | Fig. 6.1 calibration plot (10 groups of 40, n = 400), Fig. 6.2 decision curve, Tables 6.1 and 6.2 |
| II | Reading an AI study critically | Which findings support measurement quality, and which support better care? | Clinicians preparing a journal club, decision-support committees, reviewers | Required (W2 core reading) | Yes: reconstruction from published counts with stopifnot (3 checks), journal-club card as data frame | Table 7.1 tools and checks, Table 7.2 checked statement, Table 7.3 reconstructed 2 x 2 table (n = 38,455 hospitalisations, published) |
| II | Language models in clinic and science | Which statement can be checked after generation? | Clinicians, researchers, teachers and students who use or review generated text | Required (W2 core reading) | Yes: stored answers exported to data/b08-stored-answers.csv, 8 stopifnot checks | Fig. 8.1 share of false, true and other sentences at three temperatures (2,000 generated sentences each), Table 8.1 evidence card |
| III | A fracture alert in the emergency department | What does an alert mean when the prevalence changes? | Clinicians, medical students and reviewers acting on an image-analysis alert | required (W3 core reading) | Base R: fourfold counts, PPV/NPV, alerts per shift, counter-check with specificity 80%. stopifnot: 45/140 and 180/260, 32.1% and 69.2%. Download: data/b09-fracture-alert-cohorts.csv | fig-b09-icons (icon array, n = 1,000 patients per setting), fig-b09-ppv (PPV vs prevalence), tbl-b09-fourfold |
| III | Hyperspectral imaging | What separates a biological signal from illumination and device characteristics? | Doctoral researchers and preclinical teams analysing spectral images, reviewers | required (W4 core reading) | Base R plus class::knn and prcomp. stopifnot: grey-tile tolerance, pixel vs animal p-value pattern, index accuracy >= model accuracy on new animals, 50% under stale calibration, brightness-only counter-check | fig-b10-chain (measurement chain), fig-b10-index (index per animal, n = 8), tbl-b10-units |
| III | Measuring gait from video | Are two measurement methods similar enough for this rehabilitation decision? | Physiotherapists, rehabilitation and trauma clinicians, researchers comparing measurement methods | supplementary (W4 gait station) | Base R plus ggplot2. stopifnot: two hand-checked angles, step-level limits outside and 10-step limits inside tolerance, wider bias CI with patients as unit, bias shift with 30 degree yaw. Download: data/b11-gait-steps.csv | fig-b11-chain, fig-b11-perspective (error vs camera yaw), fig-b11-ba (Bland-Altman, n = 24 patients) |
| III | Spatial biology and image analysis | How does a segmentation decision change the biological conclusion? | Researchers in pathology and translational immunology, doctoral researchers, reviewers of spatial analyses | supplementary (W4 tissue station) | Base R plus ggplot2. stopifnot: design (12 patients, 36 fields, minimum distance), monotone shares, cell-level p < 0.01 but patient-level p > 0.05 at 4 micrometres, ratio test p > 0.05, edge correction raises share | fig-b12-chain, fig-b12-field (detail at 1 and 4 micrometres), fig-b12-shift (share and ratio vs expansion, n = 12 patients) |
| IV | Longitudinal biomarkers | Whose course does the estimated slope describe? | Clinical researchers, doctoral students and reviewers working with repeated laboratory values | required (core reading W5) | Yes: synthetic cohort of 36 persons (186 of 216 planned values), 21 stopifnot conditions, CSV data/b13-selenium-synthetic.csv | fig-b13-spaghetti (spaghetti plot with persons measured per visit), fig-b13-delta (tipping-point curve) |
| IV | Cluster, gradient or artefact | When does a figure become a testable subtype hypothesis? | Translational researchers, doctoral students and reviewers of patient cluster maps | supplementary (recommended reading W5) | Yes: 300 synthetic persons from one continuous gradient, 12 stopifnot conditions | fig-b14-slices (PCA map and stacked PC1 histogram), fig-b14-countercheck (gradient vs gap) |
| IV | Mechanistic models and pharmacometrics | Which sampling times make it possible to separate clearance from volume? | Clinical pharmacology trainees, doctoral students planning sampling, reviewers of PK analyses | required (core reading W6, alternative to B18) | Yes: virtual parameters only (dose 100 mg, V 20 L, CL 5 L/h), 16 stopifnot conditions | fig-b15-identifiability (curves coinciding at 4 h, normalised sensitivities) |
| IV | Bayes, small samples and the 3Rs | What uncertainty remains even when an event rate looks striking? | Preclinical researchers, doctoral students, animal welfare bodies, reviewers of small experiments | supplementary (recommended reading W5 and W6) | Yes: 10 fictitious rats (7/9 bridged), 16 stopifnot conditions | fig-b16-posteriors (priors, posteriors and intervals) |
| V | Forecasting patient flow | Is the forecast interval of an emergency department usable for next week? | ED and nursing leads planning rosters, hospital operations analysts, clinicians reading forecast dashboards | Required | Yes: synthetic negative-binomial hourly arrivals (26 weeks), forecast::snaive checked against hand computation, temporal train/test check; writes data/b17-ed-arrivals.csv | fig-b17-coverage (weekly coverage, n = 18 forecast weeks), fig-b17-surge, fig-b17-shiftmap, tbl-b17-summary |
| V | Mass casualty incidents, operating capacity and simulation | Which resource limits care in the scenario under study? | Trauma surgeons, anaesthetists and nurses writing mass casualty plans, emergency planners, operations analysts | Required (elective with B15) | Yes: simmer::add_dataframe with stored arrivals/service times, digest fingerprint of event lists; writes data/b18-event-list.csv | fig-b18-waits (wait for OR by triage and scenario, 200 replications), tbl-b18-summary |
| V | From one centre or species to the next | To which new case should the claim refer? | Doctoral researchers in preclinical imaging, clinical teams planning validations, reviewers | Required | Yes: base R hand calculation with stopifnot target values, class::knn on synthetic spectra (seed 4419, separate knn_ objects); writes data/b19-spectra.csv incl. fold columns | fig-b19-splitplan (printable split plan, n = 4 animals), fig-b19-spectra, tbl-b19-levels, tbl-b19-splitplan, tbl-b19-repeat |
| V | Pilot operation, impact and monitoring | Which observation would stop the introduction? | Clinical leads of pilots, physicians and nurses acting as human oversight, quality and patient safety officers, reviewers of early clinical evaluations | Supplementary | Yes: base R simulation (seed 2003), qbeta exact intervals, Poisson stop rule, suppression check | fig-b20-pchart (weekly night-shift alert share, n = 20 weeks), tbl-b20-sheet (monitoring sheet) |
| VI | Ethics at the bedside and in study design | Who bears the harm of a wrong decision, and who can object? | Clinicians, nurses, researchers, reviewers | Supplementary | Yes: synthetic group counts for setting A, Wilson intervals checked against prop.test(), counter-checks for setting B and larger denominators (stopifnot) | fig-b21-subgroups (error rates by group, n = 1,000 patients) |
| VI | Regulating research software and clinical AI | Which change in the intended purpose changes the regulatory question? | Research teams, clinical IT, regulatory staff, clinicians, reviewers | Required (W7) | No lab (decision exercise), timeline figure with date checks (stopifnot) | fig-b22-timeline (legal dates, status 8 November 2026) |
| VI | Data protection along the entire data flow | Who receives which data, for what purpose, and for how long? | Clinical researchers, data stewards, teachers, reviewers | Required (W7) | Yes: data-flow register with rule checks, two counter-check versions, 72-hour clock (stopifnot) | fig-b23-flow (data flow with identifiability, boundaries and deletion points, fictional) |
| VI | Information security and recovery | How does a team protect findings data, the model version and availability? | Clinical IT, research software engineers, data stewards, clinical leads | Supplementary | Yes: role-based access test, SHA-256 release check with digest, hash-chained log, restore test with three synthetic cases, two counter-checks (stopifnot) | fig-b24-threat (threat model with trust boundaries and attack surface, fictional) |
| VII | Security risks of learning and generative systems | Which instruction or input may cross a trust boundary? | clinical teams introducing an assistant or model service, research software engineers, information security officers, reviewers | supplementary | yes: simulated access service (8 synthetic requests, 3 configurations), checksum check of a model artefact with digest, minimal log, stopifnot | fig-b25-control-matrix, tbl-b25-boundaries |
| VII | Secondary use and responsible open science | Which form of scientific traceability holds up when the data are sensitive? | clinical and doctoral researchers who publish, data stewards, ethics and data access committees, reviewers | supplementary | yes: synthetic registry extract (600 records, rare group of 3), equivalence classes, homogeneity, primary and secondary suppression, data dictionary check, stopifnot | fig-b26-classes, tbl-b26-access-matrix, tbl-b26-ehds |
| VII | Model passport, study report and decision | Which evidence does an independent team need for exactly this target statement? | project groups, doctoral researchers before colloquium or submission, reviewers, decision makers | required | yes: B19 teaching case (4/4 spectra known animals, 3/6 spectra and 1/2 animals new animals, ID overlap 0), exact intervals with qbeta, passport built from the analysis, stopifnot | fig-b27-denominators, tbl-b27-reporting-paths, tbl-b27-passport, tbl-b27-decision-sheet, tbl-b27-counter-review, tbl-b27-blank-passport, tbl-b27-blank-decision |
| VII | Case atlas, glossary and transfer | From which decision do I find the right method again? | all readers starting a project, course participants in W1 and W8, supervisors | required | yes: tables stored as data, consistency checks (all 28 chapters reachable, no dangling IDs, at least 60 glossary terms), route search for two reader tasks, path extension when the target changes, stopifnot | fig-b28-atlas-map, tbl-b28-risk, tbl-b28-goal, tbl-b28-data, tbl-b28-glossary, tbl-b28-mistakes, tbl-b28-next |
Appendix B — Chapter register
R. Heller · Date: 2026-11-08 · Last changed: 2026-11-09
The register lists every chapter with its guiding question, intended readers, prerequisites, learning activity, course workshop and primary source. It is also available as a machine-readable file: editorial/chapter-register.csv.