Why symptom improvement is not evidence of slowed progression
Levodopa can improve motor function while it is present in the body, even if it does not change the rate of neuronal loss. A motor score collected during treatment therefore combines at least two processes: the underlying disease-related impairment and an acute symptomatic effect. If the score improves, that observation alone cannot tell us whether the disease trajectory changed.
The distinction matters when interpreting longitudinal Parkinson disease data. Dopamine measurements can respond to levodopa, motor scores can respond rapidly to symptomatic treatment, and neuronal-integrity measures are intended to track a slower process. These measures provide complementary information, but they are not interchangeable. As established in the earlier material on interpreting CSF, PET, and neurofilament biomarkers in progression models, a biomarker is informative about progression only to the extent that its relationship to the underlying state is understood and sufficiently stable.
A model with separate timescales
Let \(N(t)\) denote a normalized neuronal-integrity state, with larger values representing greater integrity. For a simple model, assume that integrity declines exponentially. Let \(A\) indicate assignment to a treatment hypothesized to modify progression, and let \(C(t)\) be levodopa concentration in a relevant exposure compartment. A parsimonious model is
Here \(k_0\) is the untreated progression rate and \(\delta\) is the reduction in that rate attributed to the candidate disease-modifying treatment. Both have units of inverse time. The constraint ensures that the treated integrity state does not increase under this particular model. If \(\delta=0\), the model has no disease-modifying effect; if \(\delta>0\), progression is slower in the treated group.
For a motor score \(Y(t)\) where larger values mean worse impairment, one illustrative observation model is
In this equation, \(Y_0\) is the reference motor score, \(\gamma\) converts loss of normalized integrity into motor impairment, and the concentration-dependent term represents acute symptomatic relief. \(S_{\max}\) is the maximum score reduction, \(EC_{50}\) is the concentration producing half that reduction, and \(\epsilon(t)\) represents observation error. The score, \(Y_0\), \(\gamma\), and \(S_{\max}\) must use compatible score units; \(C(t)\) and \(EC_{50}\) must use the same concentration units.
Dopamine data can help describe the rapid pharmacological response, but an assay may reflect both endogenous dopamine-related biology and levodopa exposure. A simple observation equation could be \(Z_D(t)=a_DN(t)+b_D C(t)/(EC_{50}+C(t))+\epsilon_D(t)\). A neuronal-integrity measure could be represented as \(Z_N(t)=a_NN(t)+\epsilon_N(t)\). These are illustrative links, not universal assay mechanisms. The model must be adapted to the actual biomarker, its scale, and its known response to treatment. In particular, do not count a measured dopamine response as independent confirmation of progression if it is itself directly affected by levodopa.
Deriving the washout contrast
The model gives a direct reason that washout can be informative. If levodopa exposure becomes negligible after washout, then \(C(t)\) approaches zero and the acute term vanishes:
Thus, after adequate washout, a between-group motor difference can be consistent with different underlying disease states rather than a current acute levodopa effect. It is not automatically proof of disease modification: residual drug, long-lasting symptomatic effects, differences in care or adherence, and measurement error can all complicate that interpretation.
A short derivation shows how integrity data estimate progression in this model. Taking logarithms gives \(\log N(t)=\log N_0-kt\), where \(k=k_0-\delta A\). For two measurements at \(t_1\) and \(t_2\), the rate estimate is
If integrity falls, the ratio is below one, its logarithm is negative, and the estimated rate is positive. With several time points, fitting a line to log integrity uses the full trajectory rather than only its endpoints. This derivation assumes positive measurements on a scale for which proportional decline is plausible; it does not justify taking logarithms of a noisy marker that can be zero or negative.
Worked Example: Separating a Levodopa Response From a Washout Score
Hypothetical data. A participant has a motor score of 34 points before a dose, 23 points near peak exposure, and 33 points after a planned washout. Larger scores indicate worse impairment. Suppose measured concentrations are 0, 6, and 0.1 mg/L, respectively. Use the illustrative acute effect with \(S_{\max}=15\) points and \(EC_{50}=2\) mg/L.
At peak exposure, the modelled symptomatic reduction is
The observed reduction from 34 to 23 is 11 points, close to this illustrative prediction. At the post-washout concentration of 0.1 mg/L, the predicted residual acute reduction is
The post-washout score of 33 is therefore not equivalent to a perfectly drug-free score; the model predicts a small residual acute effect. The comparison is consistent with substantial short-term symptom relief and little remaining acute effect after washout. It cannot establish that neuronal loss did or did not slow, because this single-person score series contains no neuronal-integrity trajectory or untreated comparison.
Fitting and comparing progression explanations
A practical analysis fits the information streams together while preserving their different roles. The progression parameter is informed principally by repeated neuronal-integrity measurements and, where validated, corroborating biomarkers. Dopamine and measured concentration inform acute pharmacology. Motor observations during exposure help estimate the symptomatic component; motor observations after adequate washout help assess the model’s predicted underlying impairment. A joint fit can estimate shared latent states and uncertainty, but it cannot make an invalid biomarker relationship valid.
State whether larger motor scores mean worse function, how integrity is normalized, and whether each dopamine measure is expected to respond directly to levodopa.
Use observed or model-predicted concentration and a concentration-response term. Check that predicted acute changes are compatible with measurements during dosing.
Estimate group-specific rates from repeated integrity data under a stated progression model. Compare the model with \(\delta=0\) against one allowing \(\delta>0\).
Set acute exposure to zero, or to a justified residual level, and predict motor scores and integrity measures at the observed follow-up times.
Look for systematic residual patterns, inconsistent biomarkers, or results that depend strongly on one time point or an unverified assumption.
The comparison should include the actual treatment context. For example, if levodopa is used in both groups, its acute effect may be present in both arms and still needs to be represented; the disease-modifying contrast concerns whether the progression rates differ. If treatment exposure or dose differs over time, a simple constant symptomatic term may be inadequate. Likewise, a treatment that changes biomarker production or measurement without changing neuronal integrity can appear to alter progression unless that possibility is addressed.
Worked Example: Estimating a Difference in Integrity Decline
Hypothetical data. Suppose normalized integrity is 1.00 at baseline in both groups. At years 1 and 2, the comparison group has values 0.9200 and 0.8464; the candidate-treatment group has values 0.9500 and 0.9025. Assume the exponential model is plausible over these two years and the marker measures integrity comparably in both groups.
For the comparison group, the rate from baseline to year 2 is
For the candidate-treatment group,
The estimated rate difference is \(0.08338-0.05129=0.03209\text{ year}^{-1}\), a slower decline in the candidate-treatment group under this model. At year 2, the integrity difference is \(0.9025-0.8464=0.0561\). If the motor impairment coefficient were \(\gamma=20\) points per unit integrity loss and both groups were assessed after adequate washout, the corresponding modelled motor-score difference would be \(20(0.0561)=1.122\) points, with the treated group predicted to have the lower, better score.
This calculation is evidence about the specified model, not by itself a clinical conclusion. A robust analysis would use all repeated observations, quantify uncertainty, check measurement comparability, and assess whether the exponential trend fits. The one-point follow-up prediction is a consequence of the assumed motor-integrity relationship; it is not an independently observed treatment benefit.
Worked Example: Follow-Up Predictions With Unequal Starting Integrity
Hypothetical predictions. Suppose two groups have calibrated integrity values \(N_0=0.64\) and \(N_0=0.72\) at the start of a three-year prediction, and both are assigned the same progression rate \(k=0.0372\text{ year}^{-1}\). These unequal starting values are included to illustrate why baseline integrity must be accounted for when comparing follow-up values.
The model predicts the first group's year-3 integrity as
For the second group,
The second group has the higher predicted value because it started higher, not because this calculation gives it a slower progression rate: both rates were set equal. A disease-modifying claim would require evidence that the rates differ after accounting for baseline and other relevant factors. This example also illustrates the distinction between predicting a state and estimating the process that generated its change.
Diagnostics, alternatives, and limits
A model that separates symptomatic and disease-modifying effects should be challenged with diagnostics. Plot each participant’s motor score, dopamine measure, and integrity marker over time, marking doses, concentration measurements, and washout. The model is suspect if motor residuals systematically depend on concentration, if its predicted acute response has the wrong timing, or if integrity residuals show curvature that contradicts constant exponential decline. Check whether estimated progression rates are biologically plausible and whether conclusions change materially when influential observations or uncertain biomarker mappings are considered.
Washout must be long enough for the relevant acute effect to dissipate, but the required duration depends on drug disposition and possible longer-lasting pharmacodynamic effects. A low measured plasma concentration does not guarantee that every effect-site or downstream effect has disappeared. Conversely, a persistent motor difference after washout may reflect reserve, learning, symptomatic carryover, or measurement variability as well as altered degeneration. Follow-up after treatment cessation is especially informative when the competing models make different predictions about whether a treatment-related advantage persists or converges; it still requires an explicit model of carryover and progression.
| Approach | Strength | Main limitation |
|---|---|---|
| On-treatment motor comparison alone | Directly measures function during usual treatment. | Mixes acute symptom relief with underlying progression. |
| Washout motor assessment | Reduces current acute drug contribution. | Residual or persistent effects and imperfect washout can remain. |
| Longitudinal integrity-marker comparison | Targets a slower disease-related process. | Depends on marker validity, measurement stability, and the assumed trajectory. |
| Joint model of biomarkers, exposure, and motor function | Uses complementary data and supports cross-stream predictions. | Can be hard to identify and may be sensitive to structural assumptions. |
These approaches are complementary rather than interchangeable. An analysis based only on motor scores may be simpler, but cannot cleanly separate the mechanisms without dosing or washout information. An analysis based only on an integrity biomarker may miss treatment effects on function and can mistake a direct biomarker effect for slower neuronal loss. A joint model makes assumptions explicit and enables prediction, but its apparent precision does not compensate for weak data or an incorrect structure. Even in a randomized study, a model-based difference in progression is not proof of a particular biological mechanism.
Common pitfall: “The motor score was better during levodopa treatment, so levodopa slowed neurodegeneration.” This reasoning confuses a clinical manifestation with the disease process. The correction is to estimate acute response from exposure-related observations, estimate progression from longitudinal integrity data and appropriate follow-up, and examine motor outcomes after a justified washout. A persistent off-treatment difference can support a progression explanation only if alternative explanations and measurement assumptions are addressed.
Check Your Understanding
Answer each question, then compare with the brief solutions.
- If a motor score improves during dosing but returns to its prior level after adequate washout, which model component is most directly supported?
- In the exponential integrity model, an integrity ratio over two years is 0.90. Calculate the estimated progression rate.
- Why is a higher follow-up integrity value not, by itself, evidence of a slower progression rate?
- Name one reason that a post-washout motor difference may not establish disease modification.
- What happens to the acute concentration-response term as concentration approaches zero?
Solutions. (1) The concentration-dependent symptomatic component, because the improvement tracks drug presence and diminishes after washout. (2) \(\hat{k}=-\log(0.90)/2=0.05268\text{ year}^{-1}\). (3) The groups may differ at baseline; compare estimated change or rates after accounting for baseline. (4) Residual or persistent pharmacodynamic effects, differences in reserve, or measurement variability could produce the difference. (5) It approaches zero, since \(S_{\max}C/(EC_{50}+C)\to0\) as \(C\to0\).