aiMOOC

Aging – An intervention ledger for causal scrutiny

QR-Code

Introduction

Aging – An intervention ledger for causal scrutiny is a methods-focused aiMOOC for learners in neuroscience, biostatistics, translational medicine, epidemiology, data science, and research governance. The course treats AG-T01 as a study-program identifier and designs an auditable intervention ledger for it. It does not claim that AG-T01 experiments have been performed, that any intervention works, or that any biological mechanism has been demonstrated.

Your task is to learn how to make a causal claim difficult to overstate. The ledger therefore separates three classes of statements:

Observation: A measured or documented fact, with provenance, timestamp, assay or instrument, unit, uncertainty, and quality-control status.
Model assumption: A condition needed for interpretation, identification, transport, or analysis that is not established merely by observing the data.
Proposed test: A prospective procedure intended to challenge an assumption or estimate an effect, with a frozen analysis rule and decision consequence.

The focal causal chain is deliberately broad: population → baseline impairment → intervention and comparator → exposure → cargo balance → synapse preservation → function → harm. Each link must be specified before outcomes are inspected. The ledger is versioned so that later changes remain visible rather than silently replacing earlier commitments.

Why an intervention ledger?

A conventional protocol describes what investigators plan to do. A causal intervention ledger adds explicit traceability across what was observed, what was assumed, what will be tested, and what decision rule follows. This helps you distinguish a plausible biological story from an identified causal effect.

A strong ledger also prevents a common error in aging research: treating an intermediate biomarker as if it automatically established functional benefit. Synapse-related measures can be valuable mechanistic endpoints, but a biomarker and a patient-relevant functional outcome answer different questions. National Institute on Aging research priorities explicitly emphasize integrating structural, functional, molecular, and behavioral readouts when studying synaptic and axonal degeneration.

AG-T01 Scope and Non-Claim Status

AG-T01 is defined here as a prospective causal specification object. Any field that lacks empirical data is marked UNOBSERVED. Any numerical boundary that has not yet been justified is represented by a symbolic parameter and must be frozen before confirmatory analysis.

Non-claim statement: This aiMOOC supplies a ledger design, not experimental results. Words such as “effect,” “preservation,” or “benefit” refer to estimands or success criteria unless an observation record is explicitly present.

Ledger identity and version control

Every ledger release should carry the following metadata.

Field	Required content	Audit rule
Ledger ID	AG-T01	Immutable within the study program
Version	Semantic version such as 0.1.0, 0.2.0, 1.0.0	Increment whenever a field, threshold, estimand, or test changes
Status	Draft, frozen, amended, or retired	Confirmatory analyses may use only a frozen version
Effective timestamp	Date and time with timezone	Must precede access to blinded or unblinded outcome summaries when used prospectively
Change author	Responsible person or group	Must be attributable
Change reason	Scientific, operational, safety, or correction	Free-text reason plus linked issue identifier
Supersedes	Prior ledger version	Never overwrite the historical record
Data snapshot	Dataset hash or immutable snapshot identifier	Analysis output must point to the exact snapshot
Analysis code	Version-control commit or archived package identifier	Code and ledger version must be cross-referenced

A useful rule is: no silent edits. If a threshold changes because pilot data were seen, the ledger must say so and classify the resulting analysis as exploratory unless an independent confirmatory dataset remains untouched.

Core AG-T01 Intervention Ledger

Population

Observation record: UNOBSERVED until a cohort exists. Record age range, recruitment frame, setting, sex and gender variables when relevant, key comorbidities, medication exclusions, cognitive or functional inclusion criteria, and the number screened, enrolled, randomized, treated, and analyzed.

Model assumptions: Eligibility criteria define a target population to which the effect is intended to generalize. Transport from the enrolled sample to a broader aging population requires assumptions about effect modification, selection, access to the intervention, and measurement comparability.

Proposed tests: Compare baseline distributions between the enrolled sample and the stated target population; prespecify subgroup effect-modification analyses only when scientifically motivated; evaluate positivity by checking whether each relevant covariate pattern can plausibly receive either intervention or comparator.

Ledger fields: target_population_text, eligibility_version, recruitment_frame, site_list, age_min, age_max, baseline_risk_variables, transport_covariates, positivity_flag.

Baseline impairment

Baseline impairment defines where participants begin and determines what “preservation” can mean.

Observation record: UNOBSERVED. Record the instrument, version, assessor, timing, scale direction, reliability information, clinically interpretable range, and whether the measure reflects cognition, synaptic function, everyday function, or another domain.

Model assumptions: The baseline measure captures a stable-enough construct for stratification or adjustment; practice effects and measurement error are not so large that they dominate change scores; baseline status precedes intervention exposure.

Proposed tests: Assess test-retest reliability when repeat measurements are feasible; check floor and ceiling effects; prespecify whether baseline enters the model as a continuous covariate, stratum, or both; audit differential baseline missingness.

Ledger fields: baseline_measure_id, domain, scale_range, impairment_threshold_source, baseline_window, reliability_plan, missingness_code.

Intervention

The intervention record must be reproducible without relying on brand language or biological shorthand.

Observation record: UNOBSERVED. Record formulation, manufacturing lot where applicable, route, schedule, intensity, dose modification rules, delivery personnel, adherence, co-interventions, and deviations.

Model assumptions: The intervention is sufficiently well-defined for consistency; different implementations classified under the same treatment label are not causally heterogeneous beyond acceptable limits.

Proposed tests: Conduct implementation fidelity checks; quantify adherence or delivered dose; test predefined product or delivery specifications before outcome interpretation.

Ledger fields: intervention_definition, formulation_version, schedule, route, intended_dose, delivered_dose, fidelity_metric, deviation_log.

Comparator

The comparator determines the causal contrast.

Observation record: UNOBSERVED. Specify placebo, sham, active control, usual care, delayed intervention, or another explicit condition.

Model assumptions: The comparator controls the non-specific factors relevant to the causal question, such as attention, contact time, expectation, procedure burden, or co-intervention access.

Proposed tests: Measure treatment expectancy when relevant; audit contact time and ancillary care; document contamination and cross-over.

Ledger fields: comparator_definition, masking_plan, contact_time, contamination_metric, rescue_policy, crossover_policy.

Exposure

Exposure is the participant-level quantity representing what was actually received or biologically available, distinct from random assignment.

Observation record: UNOBSERVED. Examples include delivered dose, concentration-time summary, target engagement, session completion, or another intervention-appropriate exposure measure.

Model assumptions: The exposure metric is measured with acceptable error and is temporally prior to downstream mechanistic outcomes.

Proposed tests: Validate the exposure assay or adherence metric; estimate measurement error; examine whether exposure distributions overlap sufficiently for dose-response analyses.

Ledger fields: exposure_metric, unit, assay_id, lower_limit, upper_limit, sampling_times, target_range, measurement_error_plan.

Cargo balance

Cargo balance is intentionally platform-neutral. Until AG-T01 specifies a delivery technology, it means the prespecified composition, stoichiometry, or relative abundance of intended payload components and relevant unintended or off-target cargo. It must not be treated as a measured mediator merely because it appears in the causal story.

Observation record: UNOBSERVED. Record each cargo component, measurement method, unit, normalization denominator, expected ratio or profile, batch, and uncertainty.

Model assumptions: The chosen cargo metric corresponds to the biologically relevant balance; normalization does not induce misleading compositional effects; measurement error is not strongly differential by treatment arm.

Proposed tests: Perform assay qualification; blind batch-level quality-control review; compare component-wise and compositional summaries; prespecify acceptable tolerance regions rather than choosing them after looking at functional outcomes.

Ledger fields: cargo_schema_version, component_ids, normalization_rule, balance_metric, target_region, batch_id, off_target_panel, qc_status.

Synapse preservation

Synapse preservation is a mechanistic endpoint and must be operationalized in advance.

Observation record: UNOBSERVED. Candidate measurement families include synaptic density imaging, synaptic protein biomarkers, microscopy, electrophysiology, or network-level proxies. The ledger must name the exact modality rather than treating them as interchangeable.

Model assumptions: The selected measure is sensitive to the relevant synaptic process, is not dominated by unrelated pathology, and is measured after exposure but before or alongside the functional endpoint in a causally interpretable sequence.

Proposed tests: Validate assay reliability; include negative and positive controls where appropriate; prespecify region or circuit; test robustness to alternate processing pipelines; distinguish structural preservation from functional synaptic activity.

Ledger fields: synapse_endpoint, modality, anatomical_region, timepoint, preprocessing_version, reliability_target, minimum_effect_symbol.

Function

Function is the outcome most directly connected to meaningful capability.

Observation record: UNOBSERVED. Define the primary functional domain, instrument, scale direction, assessment window, assessor masking, and whether higher or lower values are better.

Model assumptions: The measure is responsive over the study interval and interpretable in the enrolled population; practice, expectancy, and missingness do not create a spurious difference.

Proposed tests: Prespecify a primary functional endpoint; include alternate measures only as supportive outcomes; evaluate blinded assessment quality; predefine the clinically meaningful effect threshold or justify why only a standardized effect is appropriate.

Ledger fields: function_endpoint, domain, instrument_version, primary_timepoint, direction, minimum_effect_symbol, assessor_masking.

Harm

Benefit criteria are incomplete without harm criteria.

Observation record: UNOBSERVED. Record all adverse events relevant to the intervention, including severity, seriousness, relatedness, timing, reversibility, discontinuation, and deaths where applicable.

Model assumptions: Safety surveillance captures events with sufficiently similar intensity across groups; adjudication is not biased by knowledge of assignment.

Proposed tests: Establish independent or blinded adjudication when feasible; compare event rates and exposure-adjusted rates when relevant; monitor prespecified sentinel events continuously.

Ledger fields: harm_dictionary_version, sentinel_event_list, severity_scale, adjudication_plan, harm_boundary_symbol, stopping_authority.

Observation, Assumption, and Test Register

Each causal statement gets a unique ID and exactly one primary class.

ID pattern	Class	Meaning	Example	Allowed update
OBS-###	Observation	Data or documented fact	OBS-014 exposure assay concentration at week 4	New data append a new record; raw values are never rewritten
ASM-###	Model assumption	Unverified condition needed for interpretation	ASM-007 no important unmeasured mediator-outcome confounding	May be amended with reason and impact assessment
TST-###	Proposed test	Planned challenge to an observation or assumption	TST-022 blinded assay repeatability study	May change only by versioned amendment

Every test record should link to the assumption it challenges, the data required, the analysis code, the pass or fail rule, and the decision consequence.

Causal Structure and Estimands

A directed acyclic graph is a compact way to state causal assumptions. For AG-T01, a minimal conceptual graph might include assignment, received exposure, cargo balance, synapse preservation, function, harm, baseline impairment, and common causes of mediator and outcome. The graph is not evidence by itself. It is a declaration of the structure assumed by the analysis.

Primary estimands

Before analysis, define at least two estimands.

Total effect: The effect of assignment to AG-T01 versus comparator on the primary functional endpoint at the primary timepoint, using a treatment-policy or other explicitly selected intercurrent-event strategy.
Safety estimand: The contrast in a prespecified harm measure over the safety window.
Mechanistic estimand: The effect of assignment on synapse preservation, treated as distinct from mediation unless identification conditions are satisfied.

Each estimand should include the population, treatment conditions, endpoint, timepoint, handling of intercurrent events, and summary measure. ICH E9 R1 emphasizes that the estimand and the strategy for intercurrent events should be explicit and aligned with the analysis.

Intercurrent events

Examples include treatment discontinuation, rescue therapy, cross-over, death, or inability to complete functional testing. The ledger must state whether each is handled by a treatment-policy, hypothetical, composite, while-on-treatment, or principal-stratum strategy. Missing data must not be confused with a deliberate estimand strategy.

Non-Identifiable Mediation

A mediation claim such as “AG-T01 improves function because it preserves synapses” is stronger than showing that AG-T01 changes synapse measures and function.

Default ledger status: NON-IDENTIFIABLE unless the required conditions are defensible. These commonly include temporal ordering, consistency, positivity, no important unmeasured exposure-outcome confounding, no important unmeasured exposure-mediator confounding, and no important unmeasured mediator-outcome confounding. Some natural direct and indirect effects additionally require assumptions about mediator-outcome confounders affected by treatment.

Therefore:

Do not report a mediator coefficient as proof of mechanism.
Do not condition on post-treatment variables merely because they predict outcome.
Record the causal mediation estimand separately from the total treatment effect.
If assumptions are not defensible, report descriptive pathway associations or interventional analogues with clearly stated interpretation limits.
Add sensitivity analyses for mediator-outcome confounding when a mediation analysis is attempted.

The AGReMA reporting guidance recommends stating the causal model and its assumptions explicitly rather than leaving them implicit.

Missing Data and Data Quality

Missingness is a causal and statistical problem, not merely a software setting.

Observation fields: For every planned variable, record expected collection time, actual collection status, reason for missingness, whether the participant remains under follow-up, and whether missingness followed an intercurrent event.

Model assumptions: Any imputation or likelihood-based procedure carries assumptions about the distribution of unobserved values. A Missing At Random assumption should be justified using the observed history; it should not be asserted by default.

Proposed tests and sensitivity analyses: Prespecify multiple imputation or another primary method consistent with the estimand; add delta-adjustment, pattern-mixture, tipping-point, or reference-based sensitivity analyses when appropriate. Report missingness by treatment arm and endpoint.

The ledger should flag:

MD-FLAG-1: primary endpoint missingness exceeds the frozen tolerance.
MD-FLAG-2: reasons for missingness differ materially by arm.
MD-FLAG-3: post-discontinuation outcomes required by the estimand were not collected.
MD-FLAG-4: the primary conclusion changes under a prespecified plausible sensitivity range.

ICH E9 R1 recommends clinically plausible missing-data assumptions and sensitivity analyses aligned to the estimand.

Joint Success Criteria

AG-T01 should not be declared successful because one endpoint crosses a threshold. The ledger uses a conjunctive gate: all critical dimensions must pass.

Before any confirmatory outcome review, freeze the following symbols and their numeric values:

Gate	Symbol	Prespecified success condition	Why joint?
Exposure	E_min	Target exposure or target-engagement criterion is met	Without adequate exposure, downstream interpretation is weak
Cargo balance	C_tol	Cargo profile remains inside a justified tolerance region	Prevents efficacy interpretation when delivery composition is out of specification
Synapse preservation	Delta_S_min	Estimated synapse effect meets or exceeds the minimum mechanistic threshold with the required uncertainty bound	Requires evidence at the proposed biological target level
Function	Delta_F_min	Estimated functional effect meets or exceeds the minimum meaningful threshold with the required uncertainty bound	Prevents biomarker-only success
Harm	H_max	Harm remains below the prespecified unacceptable-risk boundary	Prevents benefit from overriding unacceptable toxicity
Data integrity	Q_min	Prespecified assay, completeness, masking, and protocol-quality checks pass	Prevents a nominal effect from succeeding on unreliable data

Joint success rule: SUCCESS only if G_E AND G_C AND G_S AND G_F AND G_H AND G_Q are all TRUE in the frozen primary analysis. If any gate is FALSE or unevaluable, the confirmatory status is NOT-SUCCESS. “Not-success” is not automatically proof of no biological activity; it is a decision outcome under the prespecified joint rule.

Multiplicity procedures, confidence levels, and any hierarchy between endpoints must be frozen with the thresholds. If synapse preservation is supportive rather than confirmatory, change the gate structure before unblinding and version the ledger accordingly.

Stopping Rules

Stopping rules protect participants and preserve interpretability. Boundaries must be numerical, justified, and assigned to an authorized monitoring body before enrollment or unblinded review.

Stop type	Trigger	Default action
Safety stop	Sentinel harm exceeds H_stop or a prespecified serious-event pattern is met	Pause enrollment or dosing and trigger independent review
Product or cargo stop	Cargo quality lies outside C_stop for a defined batch or sequence of batches	Quarantine affected material and suspend exposure
Exposure futility stop	Predictive or conditional probability of achieving E_min falls below P_E_stop	Stop for inadequate target exposure
Functional futility stop	Prespecified conditional-power or predictive-probability boundary is crossed	Stop or adapt only if the adaptation was prospectively defined
Data-integrity stop	Critical masking, assay, consent, randomization, or data-provenance failure occurs	Suspend confirmatory analysis until root cause is resolved
Overwhelming efficacy stop	Only if a conservative benefit boundary plus harm conditions is crossed	Independent monitoring review; do not stop solely on a noisy biomarker

Do not invent numerical boundaries after interim data are seen. If boundaries are still unknown, the ledger should state TO BE JUSTIFIED AND FROZEN, not fill them with convenient values.

Versioned Ledger Template

A machine-readable record can be implemented in a database, spreadsheet, JSON schema, or validated electronic data-capture system. The minimum human-readable specification is:

Field	Example placeholder	Type	Required
record_id	OBS-001	String	Yes
ledger_version	0.1.0	String	Yes
object	synapse_preservation	Controlled term	Yes
class	observation	observation assumption proposed_test	Yes
status	unobserved	Controlled term	Yes
value	NA	Typed value	When observed
unit	platform specific	Controlled unit	When observed
timepoint	primary mechanistic timepoint	Time window	Yes
provenance	dataset and assay identifiers	Reference	Yes
uncertainty	confidence or posterior interval	Structured field	For estimates
qc_status	pending	pass fail pending	Yes
linked_assumptions	ASM-007	List	As applicable
linked_tests	TST-022	List	As applicable
decision_rule	symbolic until frozen	Expression	For decision records
amendment_reason	NA	Text	If changed

Audit trail events

Every material event should append, rather than overwrite, a row containing event_id, actor, timestamp, old_value_hash, new_value_hash, reason, approval_status, and affected analysis outputs. This enables an auditor to reconstruct what was known, assumed, and planned at each point in time.

Worked Hypothetical Example

The following example uses symbolic values only.

OBS-101: Baseline impairment score is planned but no participant data exist. Status = UNOBSERVED.

ASM-101: The chosen baseline instrument is sensitive to meaningful change over the AG-T01 follow-up interval. Status = UNTESTED.

TST-101: In a pilot dataset separate from confirmatory analysis, estimate test-retest reliability and floor or ceiling effects using a frozen script. Decision consequence = retain, revise, or replace the instrument before confirmatory enrollment.

ASM-202: The synapse measure lies on a pathway that may mediate functional effects. Status = NON-IDENTIFIED FOR MEDIATION.

TST-202: Collect exposure before the mediator and mediator before the functional endpoint, measure prespecified confounders, publish the DAG, and run a sensitivity analysis. Decision consequence = mediation remains exploratory unless identification assumptions are defensible.

This illustrates the key discipline: a proposed mechanism becomes a list of observable requirements and falsifiable assumptions, not a conclusion.

External Evidence and Methodological Anchors

The National Institute on Aging highlights the need to combine molecular, structural, functional, and behavioral measures in research on synaptic and axonal degeneration. It also notes methodological challenges in long-horizon prevention trials, including target selection, biomarkers, cognitive outcomes, trial design, and causal evidence.

Key methodological anchors for this ledger include ICH E9 R1 for estimands and missing data, AGReMA for transparent mediation reporting, and causal DAGs for stating identification assumptions.

Selected References

ICH E9 R1 Addendum on Estimands and Sensitivity Analysis: Framework for defining treatment effects, intercurrent-event strategies, and sensitivity analyses.
AGReMA Statement: Reporting guidance for mediation analyses, including explicit causal assumptions.
National Institute on Aging workshop on synaptic and axonal degeneration: Current research priorities linking molecular, structural, functional, and behavioral measures.
National Institute on Aging workshop on early prevention: Discussion of biomarkers, cognitive outcomes, informative trial designs, and causal evidence.

Interactive Tasks

Quiz: Test Your Knowledge

Diesen aiMOOC als Battle spielen

What is the primary purpose of the AG-T01 ledger? (To separate observations assumptions tests and decision rules) (!To prove that AG-T01 already works) (!To replace all experimental data) (!To guarantee a causal mechanism)

Which item is an observation? (A recorded assay value with provenance) (!A belief that no confounding exists) (!A proposed future reliability study) (!A biological story without measurement)

What should happen when a confirmatory threshold changes? (The change should create a versioned amendment) (!The old threshold should be deleted) (!The change should remain undocumented) (!Only the final threshold should be stored)

Why is cargo balance kept platform neutral? (The delivery technology has not been specified) (!Cargo balance is never measurable) (!Cargo balance is identical to function) (!Cargo balance replaces exposure)

What does non identifiable mediation mean here? (The causal indirect effect is not justified from current assumptions and data) (!The mediator was proven to cause the outcome) (!The total effect cannot be estimated in any design) (!The comparator must be removed)

Which endpoint prevents biomarker only success? (The functional endpoint) (!The batch identifier) (!The version number) (!The recruitment log)

How should missing post discontinuation data be treated when the estimand requires them? (As missing data requiring justified assumptions and sensitivity analysis) (!As automatically irrelevant data) (!As proof of treatment failure) (!As a reason to delete the participant)

What is a conjunctive success rule? (All critical gates must pass) (!Any single gate may pass) (!Only the synapse gate matters) (!Only the safety gate matters)

When should stopping boundaries be frozen? (Before unblinded interim outcome review) (!After the desired result appears) (!After the final manuscript is written) (!Only if a safety event occurs)

What does an auditable record preserve? (The history of data assumptions tests and amendments) (!Only the latest edited value) (!Only statistically significant outcomes) (!Only favorable safety findings)

Memory Game

Estimand	Precisely defined treatment effect target
Provenance	Traceable origin of a recorded datum
Positivity	Availability of relevant treatment alternatives across covariate patterns
Mediator	Post exposure variable on a hypothesized causal pathway
Intercurrent	Event after treatment initiation that affects interpretation or measurement
Versioning	Preservation of historical changes to the specification

Drag and Drop

Match the correct terms.	Topic
Measured fact with traceable source	Observation
Condition required for causal interpretation	Model assumption
Prospective challenge with a decision consequence	Proposed test
Outcome threshold system requiring every gate to pass	Joint success
Rule that can pause or terminate the study	Stopping criterion

...

Crossword Puzzle

Ledger	What record preserves observations assumptions tests and amendments?
Synapse	What neuronal junction is a mechanistic focus of preservation?
Exposure	What term describes what was actually received or biologically available?
Comparator	What condition defines the causal contrast with the intervention?
Mediation	What analysis studies a pathway through an intermediate variable?
Provenance	What term means the traceable origin of data?

LearningApps

Cloze Text

Open-Ended Tasks

Easy

Ledger field map: Create a one-page diagram showing the nine core AG-T01 objects from population through harm and label each as an object rather than a result.
Observation sorting: Write ten example statements and classify each as observation, model assumption, or proposed test, then explain two borderline cases.
Audit vocabulary: Build a glossary of fifteen ledger terms and illustrate each with a short aging-research example.
Synapse explainer: Create a labeled image or two-minute video explaining why synapse preservation is mechanistically interesting but not identical to functional benefit.

Standard

Version history exercise: Draft ledger versions 0.1.0 and 0.2.0, change one endpoint definition, and write the amendment reason and analysis consequence.
DAG workshop: Draw a causal graph linking assignment, exposure, cargo balance, synapse preservation, function, harm, and at least three baseline causes, then identify adjustment errors to avoid.
Missing data audit: Design a missingness codebook and show how discontinuation, death, missed visits, and technical assay failure would be represented differently.
Comparator interview: Interview a researcher or clinician about what a credible active or sham comparator must control in an aging intervention and summarize the design implications.

Advanced

Estimand specification: Write a complete primary functional estimand for AG-T01 including population, treatment conditions, endpoint, timepoint, intercurrent-event strategy, and summary measure.
Mediation challenge: Produce a memo listing every assumption required for a proposed synapse-mediated functional effect and identify which assumptions are empirically testable, partially testable, or fundamentally untestable.
Joint gate simulation: Simulate hypothetical gate outcomes for exposure, cargo balance, synapse preservation, function, harm, and data quality, then show how a conjunctive rule differs from a single-endpoint rule.
Independent audit package: Assemble a mock release containing the frozen ledger, change log, data dictionary, analysis plan, synthetic dataset, code hash, and decision log so another learner can reproduce the decision status.

Learning Assessment

Causal specification review: Given a fictional AG-T01 protocol, identify at least five statements that wrongly mix evidence with assumptions and rewrite them as ledger records.
Threshold justification: Propose a principled process for choosing Delta_S_min, Delta_F_min, H_max, and C_tol without using confirmatory outcome data.
Intercurrent event reasoning: Compare two valid estimand strategies for treatment discontinuation and explain how each changes the scientific question.
Mediation critique: Evaluate a hypothetical claim that synapse change mediates functional improvement and identify the assumptions that would be needed before causal language is justified.
Audit reconstruction: Reconstruct which ledger version governed a hypothetical analysis after three amendments and explain whether the resulting evidence is confirmatory or exploratory.

Evidence of Learning

Evidence of learning should include both knowledge and products. You should be able to explain the difference between an observation, assumption, and proposed test; define an estimand and an intercurrent-event strategy; describe why mediation may be non-identifiable; and explain why missing data assumptions belong in the causal specification.

Strong performance is shown by an auditable AG-T01 ledger with version history, operational definitions, explicit provenance, a DAG, a missing-data plan, symbolic or justified frozen thresholds, joint success gates, stopping rules, and a clear non-claim statement. Transfer is demonstrated when you can apply the same ledger logic to a different intervention, biomarker, or aging-related functional outcome without importing AG-T01-specific assumptions.

OERs on the Topic

Useful open resources include the National Institute on Aging materials on cognitive health, AD and ADRD intervention research, and synaptic and axonal degeneration; the ICH E9 R1 estimand guideline; the AGReMA mediation reporting guideline; and openly available educational material on directed acyclic graphs.

Linked Learning Areas

Aging – An intervention ledger for causal scrutiny

aiMOOC Projects

Inhalte werden geladen ...

Mediathek wird aus dem Wiki geladen ...

Frag mich zum aktuellen Lerninhalt oder wähle eine kurze Hilfe.

Textmodus bereit

Sprache