stats4PT Classroom presentation
1 / 16

Clinical Inquiry I · 2026

Causal Models

Representing population knowledge and reasoning about an individual patient

From what we know to what we think is happening

Download PDF

Opening problem

Same outcome. Same cause?

Patient A Reduced six-minute walk distance

Marked quadriceps weakness. Stable cardiovascular response. Preserved balance.

same measured result 6MWD ↓
Patient B Reduced six-minute walk distance

Adequate isolated strength. Early dyspnea. Abnormal exertional response.

Would you expect the same explanation or the same intervention?

Review of the reading

ClaimsAn association is not, by itself, evidence of causation.
ConditionsCauses operate through mechanisms and within contexts.
RealityWhat we observe does not exhaust what occurs or what can generate events.
ModelsEvery analysis presupposes some account of the causal structure.

Back to the opening problem: the same 6MWD invites different counterfactuals.

Patient A

Had quadriceps weakness been absent, would the 6MWD still have been reduced?

Patient B

Had exertional dyspnea been absent, would the 6MWD still have been reduced?

Same intervention?

Had each patient received strengthening, would each patient have improved?

A counterfactual evaluates a specified causal contrast. The observed outcome alone does not tell us which contrast is clinically relevant.

The population-to-patient problem

Which causes matter?

Sample → population

Research builds the model

Accumulating studies slowly expand and revise how well it represents population knowledge.

buildsrevises
Empiricalobserved events Actualoperative mechanisms Realgenerative mechanisms
Representative causal model of current knowledge
usesfindings may revise ←
Population → patient

Clinical reasoning uses the model

Patient evidence constrains which represented mechanisms could be operative now.

Still unknown: which causal configuration is operative in this patient?

Both sides can modify the same model. Research primarily builds it; clinical reasoning primarily uses it.

Today's next step

From observations to causal diagrams to patient-specific explanation

Population knowledge

Models4PT

Represents causal claims together with mechanisms, evidence, context, provenance, uncertainty, and disagreement.

Patient reasoning

Clinical instantiation

Uses generic causal knowledge and current patient evidence to construct a provisional patient-specific explanation.

The causal model represents population knowledge that constrains patient reasoning. It does not determine the answer.

Historical foundation

Clinical inquiry diagram

How should knowledge move between research and practice?

Clinical inquiry curriculum diagram showing facts and observations moving by induction to universals, causal models connecting universals with particulars, and deduction and abduction linking particulars with practice, all within overlapping knowledge-development and practice domains.
Plymouth State DPT clinical inquiry curriculum materials, 2015. Image provided by the course author.

Reading the 2015 logic

Clinical inquiry moves between particulars and universals

particularsobservations
→
scientific inquiryresearch + statistics
→
universalspopulation causal knowledge
→
clinical reasoningthis patient

Inductionparticular observations → general claims

Deductiongeneral knowledge → expected findings

Abductionfindings → best current explanation

Practice to researchclinical experience→patterns + anomalies→candidate claims + questions→structured research→population knowledge

A clinical encounter may be structured for patient care. Clinical experience as a whole is not structured to support population inference.

Generic causal model

A model is a structured set of claims

An epistemological representation of the causal structure believed to characterize a class of systems or patients.

Adapted from Collins, 2026 preprint

Nodes
variables or concepts selected for the purpose
Arrows
candidate claims about causal direction or influence
Paths
possible mechanisms and competing explanations
Graph form
cyclic models retain feedback paths; acyclic models prohibit return paths for a defined, often time-ordered analysis
Boundary
what the representation includes and omits

Critical realist stratification

Causal models represent knowledge across scales

muscle function→walking capacity
A model within muscle function
excitationregulationstructureenergetics
tension
A model within walking capacity
propulsionbalanceoxygen transportpacing
sustained walking

A node at one scale can open into another causal model. At every scale, the model is a fallible representation of what we know, not the generative system itself.

Developmental teaching model

Several pathways can limit walking

Draft
Revised draft Model 4D causal DAG for heart failure and neuromuscular electrical stimulation Eighteen variables connected by twenty-nine directed edges. Aerobic training causes balance, cardiac output, mechanical efficiency, and muscle function. Neuromuscular electrical stimulation and strength training cause muscle function. Muscle function causes arteriovenous oxygen difference, anaerobic threshold, and balance. Arteriovenous oxygen difference and cardiac output cause VO2 max. VO2 max causes anaerobic threshold and biological or physiological status. Anaerobic threshold, balance, gait speed, and mechanical efficiency cause six-minute walk distance; balance also causes gait speed. Six-minute walk distance causes functional status. Biological or physiological status causes functional and symptom status. Social factors cause functional status, health-related quality of life, and health perception. Symptom status causes functional status, health-related quality of life, and health perception. Functional status and health perception cause health-related quality of life. AerobicTraining NMES StrengthTraining Social Cardiac Output Muscle Function Balance MechanicalEfficiency (a-v)O₂ VO₂ max Gait Speed Bio/PhysioStatus AnaerobicThreshold Symptom Status 6MWD Functional Status HealthPerception HRQOL
Model 4D, revised · 18 variables · 29 proposed causal claims

Back to Patients A and B: Their findings direct attention to different regions of the same population causal model. They do not yet identify the operative pathway.

Population knowledge representation

The graph shows the claim, not why we should trust it

muscle function→6MWD

MeaningWhat precisely does each variable represent?

MechanismWhat lower-scale pathways does this arrow collapse?

EvidenceWhat observations or studies support the arrow?

ScopeFor whom, when, and under what conditions?

ProvenanceWhere did the claim come from, and who revised it?

UncertaintyHow strong, disputed, or incomplete is the claim?

Models4PT connects causal structure to meaning, mechanisms, evidence, provenance, context, and uncertainty.

The transition to patient reasoning

The population model informs but does not identify a patient's causal configuration

generic causal model
muscular
muscle function→6MWD
cardiovascular
cardiac output→6MWD
balance / gait
balance→6MWD
current patient evidencefindings · context · response over time
Which mechanisms may be operative?provisional patient-specific explanation

Population knowledge defines scientifically plausible possibilities. Patient evidence constrains which may explain this patient, here and now.

Patient-specific representation

Clinical instantiation

Generic causal knowledge and current patient evidence jointly constrain a provisional model of what may be happening in this patient now.

generic model + evidence now → ℐ(M, Et) = Mp,t

In words: Applying clinical instantiation to generic causal knowledge and the evidence currently available for a patient produces a provisional patient-specific causal representation. We must understand this process because it's what we're trying to teach future clinicians!

ℐconstructs a patient-specific representation

do(X)represents an intervention within a specified causal model

Clinical instantiation is a potentially composite process, not a settled mathematical operator or automated answer.

Small group activity

Build two provisional models

Fictional cases

Use the population causal model from slide 10 to construct a provisional explanation for each patient.

Patient A · difficulty attaining
  • Marked quadriceps weakness
  • Stable cardiovascular response
  • Preserved balance
  • Cannot generate enough force to rise and initiate walking
Patient B · difficulty sustaining
  • Adequate isolated strength
  • Early dyspnea
  • Abnormal exertional response
  • Preserved gait mechanics
  1. Which path is currently foregrounded?
  2. What competing explanation remains plausible?
  3. What one observation would discriminate between them?
  4. Would strength training or NMES target the represented limitation?

Debrief · new evidence

A useful model remains provisional and revisable

Patient A

New finding: repeated contractions fatigue rapidly despite improved initial force.

Revision: add a sustain/endurance hypothesis; do not discard the strength finding.

Patient B

New finding: exertional dyspnea coincides with new pulmonary crackles and an S3.

Revision: increase concern for a disease-specific limitation and reconsider intervention priorities.

Return to the opening problem: The same reduced 6MWD supports different provisional explanations. For Patient A, muscle function appears most relevant, with new evidence that endurance may also limit performance. For Patient B, the findings increase concern for a disease-specific exertional limitation. The model guides different next questions and intervention priorities; it does not provide a certain final answer.

evidence→instantiation→reasoning→action→new evidence

Educational cases only. The findings illustrate model revision, not patient-specific recommendations.

A long journey toward greater clarity

Represent population knowledge. Reason about the particular.

scientific inquirystats4PT + Physiolog
→
generic causal modelModels4PT
+
current patient evidencefindings · context · response
→
clinical instantiationℐ(M, Et)
→
provisional patient modelMp,t
→
clinical reasoningsupported by CIE

Clinical inquiry questions · observations · action · new evidence · revision

Causal models organize population knowledge; they remain fallible and revisable.

Clinical instantiation uses current evidence to construct a provisional patient-specific explanation.

Clinical inquiry continually tests and revises that explanation through questions, observations, and action.