Research model · For review
Illness, experience and intervention
Illness, perception and what someone does about them can change one another. A course enters this system by offering explanations, skills, practices and rules for noticing progress. Observation and participation become part of the feedback.
This working model keeps ongoing bodily processes, learned responses, ordinary care and course participation in play together. Seven primary studies constrain parts of it. None validates the whole system or supplies its coefficients.
Follow a possible feedback path
Select a loop to emphasize it within the whole map. Each block and arrow opens its meaning and source limits. These are candidate dependencies: their direction, size, delay and even presence must be specified for a particular condition and intervention.
A proposed family of time-ordered coupled models, not a fitted universal model. Edges include operational choices, measurement dependencies and candidate biological effects; inspect the stated role in each detail. Candidate effects may be absent, delayed or context-dependent. No gain, efficacy coefficient or complete causal mechanism is supplied by this network.
Candidate dependencies for study. Selecting this loop does not assert that it is present, dominant, reinforcing or beneficial in every condition.
Color and line style identify a connection’s role, not its evidence strength.
Selected connection 9: Intended choices shape actual exposure
Choices within real constraints → Actual practices and exposure
Highlighted: Practice and adjustment. The rest of the map stays visible.
Swipe the map sideways, or open the same blocks and connections as text below. Select a block or numbered connection to inspect it.
Read the same blocks and connections as text
Blocks
Connections
Read the highlighted loop’s connections
Connection 9 · Candidate research relationship
Intended choices shape actual exposure
Choices within real constraints → Actual practices and exposure
Operational delivery dependency: intentions are translated into actual exposure under capacity, access, burden and adherence constraints. Assignment and received dose are not equivalent; no optimal policy is assumed.
Independent evidence and its scope
An explicit modeling proposal; no direct empirical warrant is assigned here. Its purpose and assumptions are described above and in the model specification.
A model that evolves over time
A hybrid state-space model fits this job: bodily, learned and experiential states evolve, while lessons, decisions and course stages change at particular times. The participant acts on incomplete information. Their observation habits can also change the state they are trying to understand.
Variables and the order of events
- P · Bodily processes
- Condition-specific pathology, tissue damage, metabolic constraints and any justified ongoing biological input.
- R · Regulation
- Specified autonomic, endocrine or immune regulatory variables. Neither safety nor dysregulation is treated as one measurable quantity.
- L, B, K · Learning, beliefs and skills
- Learned response dispositions, explicitly held explanations, and the ability to perform a practice are different states.
- A, Q · Attention and burden
- What is attended to, and accumulated fatigue, time, cost and opportunity constraints.
- s · Course participation
- Discrete stage, available material, pausing or leaving. Advancing is not defined as recovery.
- u, m, v · Enacted inputs
- Actual practices, activity and care; monitoring actions; and explanatory material encountered. Offered and enacted doses differ.
- E, Y · Experience and observation
- Felt symptoms differ from reports, assays and recorded behavior. Changes in either deserve explanation.
X = (P, R, L, B, K, A, Q)
1. m_t ~ monitoring policy(history, experience, stage)
2. X_t^m = M(X_t, m_t, context) // observation can act
3. Y_t ~ h(X_t^m, experience, instrument)
4. u_t ~ action policy(returned information, capacity,
beliefs, skills, available choices)
5. X_(t+1) ~ F(X_t^m, u_t, material, inputs, prior load, delay)
6. s_(t+1) ~ progression policy(information available then,
skills, invitations, resources)
The new state generates later experience and later observations.
Only a returned observation can inform the next decision.These symbols define a family of models, not a fitted equation. Functions may omit unsupported paths. Their parameters, noise, missing observations and time delays still need specification. The full model separates symptom reports, function, disease measures and burden; it also distinguishes collection time from the time a result becomes available.
A Markov model can be one implementation after the state includes the relevant memory, skills and delayed inputs. A diagram of symptoms alone generally loses that history. An attractor or a “reset” would be a further hypothesis requiring a measurable restoring tendency or durable change in response dynamics.
Read the full mathematical specification and identification limits →What the seven studies contribute
The comparison supplies examples of physiological coupling, persistent inputs, delayed bodily effects and different kinds of intervention outcomes. It also tells us which arrows remain untested. The full comparison records populations, designs, findings and reading limits.
Bartsch 2015: network physiology
Measured coordination among physiological signals in healthy sleep. This constrains how to describe changing coupling; it identifies neither a unique causal law nor chronic-illness treatment effects.
Primary source ↗Dobreva 2021: inflammation, pain and cardiovascular dynamics
Fitted and independently checked acute endotoxin model. Nonlinear couplings are specified model structure; sustained pathological-input and treatment scenarios are simulations.
Primary source ↗Relouw 2024: acute and prolonged inflammatory input
Validation across endotoxin doses and a three-hour infusion. Infection inputs were selected to produce plausible ranges, not measured clinical infection trajectories.
Primary source ↗Queirazza 2024: inflammation and learning-related neural activity
Randomized acute inflammatory manipulation changed selected task-related neural signatures. Behavioral performance and fitted behavioral parameters did not significantly differ. No inflammation–belief maintenance loop was demonstrated.
Primary source ↗Appelman 2024: tissue and post-exertional changes
Long-COVID cohort with PEM and muscle/metabolic observations. Keeps tissue constraints and delayed effects visible; it does not estimate a nonlinear coupling law or test retraining.
Primary source ↗Pradhan 2007: mindfulness during rheumatoid arthritis
Small randomized package trial with continued usual care. Later psychological benefit without demonstrated disease-activity benefit; not a mechanism or component-effect estimate.
Primary source ↗Tesser 2025/2026: implanted vagus stimulation in active RA
Sham-controlled device trial with a clinical response difference during active autoimmune illness. Its device exposure is not equivalent to breathing, felt safety or commercial retraining.
Primary source ↗What this model should let us test
Begin with a small, condition-specific model. These contrasts can coexist; they are not competing worldviews that someone must choose between.
Helpful experience with or without disease change
Allow a behavioral practice to change distress, participation or symptoms while disease remains active. Compare that with an additional disease effect, retaining care and activity as possible routes. The outcomes require separate measurements; a benefit does not establish its advertised mediator.
Temporary response or durable learned change
Compare effects that track current practice with changes that persist on later, unpracticed probes. Improvement alone cannot distinguish learning from concurrent care, reduced exposure or natural fluctuation.
Observation that informs, changes or burdens
Compare feasible cadences, wording and feedback conditions. A difference may arise from attention, reporting, better decisions or burden. Do not define every change as symptom reinforcement.
State-dependent interactions or simpler input effects
Test whether a specified interaction, such as practice dose with inflammatory state, predicts held-out trajectories better than additive effects and delays. A complex diagram does not earn a nonlinear model; the extra structure must improve a prespecified test.
For rheumatoid arthritis, symptom distress and inflammatory disease measures can diverge. For long COVID with post-exertional malaise, actual activity and delayed recovery must remain visible. Neither example provides a rule for overriding symptoms or selecting a treatment. A useful next study needs a specific population, ingredient, outcomes, lag windows and rival predictions.