Representing Disease Activity as a Mechanistic State

Why Represent Disease Activity as a State?

A cytokine concentration, a tender-joint count, and the extent of tissue injury can all provide information about inflammatory disease, but they describe different things. A cytokine assay measures a molecular concentration in a sampled compartment. A clinical observation records a manifestation, which may reflect several biological processes at once. Tissue inflammation and accumulated damage are underlying biological conditions that evolve over time. A mechanistic model therefore benefits from representing those conditions explicitly rather than treating any one measurement as disease activity itself.

Here, disease activity is represented by a latent state vector with two components: active tissue inflammation and accumulated tissue damage. “Latent” means that the components are not necessarily observed directly; their values are represented within the model and may be inferred from measurements. The representation does not assert that all diseases have exactly two relevant states. It is a compact starting point that distinguishes ongoing inflammatory activity from its more persistent tissue consequences.

For the examples, \(I(t)\) is a dimensionless burden scaled to a chosen reference, while \(D(t)\) is expressed in normalized damage units. The scale must be defined for each model application; these values are not universal clinical units. Keeping the components separate matters: ongoing inflammation can fall while damage remains, and similar amounts of damage can coexist with different current inflammatory activity.

A Minimal Mechanistic State Model

A simple model can describe inflammation as the balance between inflammatory drive and resolution, and damage as the balance between inflammation-mediated injury and repair. Let \(q(t)\) be the net inflammatory drive entering the tissue state, in inflammatory-burden units per day. Let \(k_r\) be the first-order resolution rate, \(k_d\) the damage-generation rate per unit of inflammation, and \(k_p\) the first-order repair rate. The model is

The parameters \(k_r\) and \(k_p\) have units of per day. \(k_d\) has units of damage units per inflammatory-burden unit per day. Thus, both terms in each equation have compatible units. The drive \(q(t)\) summarizes the net pro-inflammatory input represented by the model; it is not automatically equal to a measured cytokine concentration. In a more detailed network, cytokine and cellular processes can contribute to \(q(t)\), and feedback can make those processes depend on tissue state. The earlier material on cytokine feedback and tissue–circulation coupling motivates such connections without making an assay identical to a state.

This minimal model assumes that the represented tissue is sufficiently homogeneous to use one inflammation state and one damage state; parameters are constant over the interval considered; \(q(t)\) is nonnegative; and resolution, damage generation, and repair are adequately approximated as first-order processes. It also assumes nonnegative initial states. Under these conditions, the equations preserve nonnegative \(I(t)\) and \(D(t)\). These are modeling assumptions, not universal biological laws. If inflammation saturates, damage is irreversible, or distinct tissue regions behave differently, the state structure or equations must change.

Steady State and Stability

When the effective inflammatory drive is constant at \(q\), a steady state is a condition in which both state derivatives are zero. Setting the right-hand sides to zero gives the equilibrium values directly.

For positive \(k_r\) and \(k_p\), the equilibrium is stable in this model. To see this, first note that for constant \(q\), the inflammation equation is a first-order linear equation. Multiplying it by the integrating factor \(e^{k_rt}\) and integrating gives

Since \(k_r>0\), the exponential term tends to zero, so \(I(t)\) tends to \(I^*=q/k_r\). The damage equation is also a stable first-order equation driven by \(I(t)\): after \(I(t)\) approaches \(I^*\), its input approaches the constant \(k_dI^*\), and damage tends to \(D^*=k_dI^*/k_p\) because \(k_p>0\). This establishes convergence for the stated model. It does not establish that the equilibrium is biologically realistic for every disease or that the system reaches it quickly.

States, Assays, and Clinical Observations

A mechanistic model distinguishes the state from the variables used to inform it. A measured cytokine \(C_k(t)\) is an assay result for a particular molecule, compartment, sampling time, and measurement procedure. It can contribute to inflammatory drive, be produced as part of inflammation, or reflect both processes. A clinical observation \(Y_j(t)\), such as a sign or symptom, is an observable manifestation. In general, it can depend on inflammation, damage, other biological factors, and measurement error.

The functions \(h_k\) and \(g_j\) represent observation relationships, and the \(\epsilon\) terms represent measurement error or unexplained variation. These relationships need not be one-to-one: different states can yield similar observations, and the same state can yield different observations if other factors vary. A measured cytokine may also be modeled as a driver, an output, or part of a feedback loop, depending on the biological model. The essential distinction is that the assay is a measurement of a molecule, while \(I\) and \(D\) are modeled tissue conditions.

Diagnostics, Alternatives, and Limitations

The state representation is useful when a model needs to keep active inflammatory processes separate from longer-lasting tissue consequences. It also makes assumptions visible: for example, the model says that damage is generated in proportion to \(I(t)\) and repaired at a rate proportional to \(D(t)\). An alternative is to analyze molecular assays or clinical observations directly without introducing latent states. That can be suitable when the objective is to describe those measurements and no mechanistic interpretation is needed. A single scalar summary is another possible representation, but it can obscure whether a change reflects inflammation, damage, or both. Defining the state vector does not itself prescribe how to combine its components into a clinical score.

Several warning signs call for checking the model rather than forcing an interpretation. Persistent disagreement between predicted and observed patterns may indicate that the observation relationship is inadequate, an important state is missing, or the assumption of constant rates is unsuitable. Estimated negative inflammation or damage values conflict with the intended nonnegative state definition and may signal weak data support, an unsuitable model, or a poorly specified estimation procedure. If only one observation type is available, its information may not distinguish \(I\) from \(D\), as the worked example showed. Model assumptions and available measurements determine what can be inferred; a mechanistic label alone does not guarantee that the states are identifiable.

A further limitation is aggregation. A single \(I(t)\) and \(D(t)\) can conceal spatially distinct tissue regions or cell populations. Linear first-order kinetics may fail when pathways saturate, thresholds matter, or repair capacity changes over time. The model also does not establish causation from an observed association: a cytokine and a state can covary because of feedback or shared drivers. These limitations are reasons to refine or test a model against appropriate biological evidence, not reasons to equate a biomarker with the state.

Common Mistake

Wrong reasoning: “The cytokine concentration fell, so disease activity and tissue damage must both have fallen.” A cytokine assay is not a direct measurement of either state component; it may be an input or output of a larger network, and the damage state can persist after inflammatory activity changes.

Correction: Specify the state variables and the relationships linking them to assays and clinical observations. Interpret a change in a measurement as evidence about the state only through those relationships, while allowing the inflammation and damage components to evolve differently.