🧬 biology

Observable Performance Does Not Fully Reflect System Organization: A Multi-Level Analysis of Gait Dynamics Under Occlusal Constraint

This study demonstrates that in adaptive neuromechanical systems, such as gait under occlusal constraints in a Parkinson's patient, comparable observable performance metrics can mask fundamentally different underlying system organizations, revealing the limitations of relying solely on aggregated linear measures.

Original authors: Jacques Raynal, Pierre Slangen, Jacques Margerit

Published 2026-05-04

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jacques Raynal, Pierre Slangen, Jacques Margerit

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Idea: The "Look" vs. The "Engine"

Imagine you are looking at two cars driving down the highway at the exact same speed, 60 mph. To a casual observer, they look identical. They are performing the same task.

However, if you pop the hood, you might find that Car A is running on a perfectly tuned V8 engine, while Car B is running on a sputtering four-cylinder engine that is being pushed to its absolute limit just to keep up.

This paper argues that in human movement (specifically walking), we often make the mistake of judging the "engine" (how the body is organized) just by looking at the "speedometer" (how fast or well someone walks).

The researchers found that two different walking conditions can look exactly the same on the outside (the speedometer), but the body's internal "engine" is actually working in completely different ways.

The Experiment: Changing the "Jaw Height"

To test this, the researchers looked at one person with Parkinson's disease. They treated the Vertical Dimension of Occlusion (VDO) as a variable.

What is VDO? Think of it as the "height" of your bite when your teeth touch.
The Setup: They put the participant in different "bite" scenarios:
- Normal bite.
- Teeth clenched tight.
- Mouth open.
- Bite raised slightly higher (by 2.5 or 3 degrees).
- Bite raised higher with the jaw pushed forward.

They asked: If we change the height of the bite, does the way the person walks change?

The Three Levels of Analysis

The researchers didn't just look at one thing; they looked at the data through three different "lenses" or levels.

Level 1: The Scorecard (What we can easily see)

This is the "Speedometer." They calculated a single number (a score) based on walking speed, step length, and balance.

The Result: When they compared the different bite heights, some conditions produced identical scores. For example, raising the bite by 2.5 degrees and raising it by 3 degrees looked exactly the same on the scorecard.
The Problem: If you only looked at this score, you would think the body was reacting the same way to both changes.

Level 2: The Dance Floor (How the body moves over time)

This level looks at the pattern of movement, not just the final score. Imagine watching a dancer. Two dancers might end up in the same spot on the stage, but one got there by gliding smoothly, while the other jumped frantically.

The Result: Even when the "Scorecard" (Level 1) was the same, the "dance" (the path the body took through space) was different. The body was organizing its movements differently to achieve the same result.

Level 3: The Hidden Map (The "Secret Code")

This is the most complex part. The researchers used a computer technique (called unsupervised embedding) to compress all the complex data into a "hidden map" (latent space). Think of this as a secret code that reveals the true identity of the movement.

The Big Discovery: On this hidden map, the two conditions that looked identical on the Scorecard (Level 1) were actually far apart. They were in completely different neighborhoods.
The Meaning: This proves that the body was using two completely different internal strategies to get the same walking result.

The "Clinical Undecidability" Trap

The paper introduces a concept called "Clinical Undecidability."

Imagine a doctor trying to fix a patient's bite. The doctor looks at the walking score and sees that Condition A and Condition B both look "good."

The Trap: The doctor might think, "It doesn't matter which one I pick; they are the same."
The Reality: The paper says, "Actually, they are totally different inside." If you pick the wrong one, you might be forcing the patient's body to use a "sputtering engine" strategy that is less stable or efficient, even though the score looks fine.

Because the scores look the same, you cannot decide which condition is better just by looking at the numbers. You need to look deeper.

What the Paper Does NOT Say

It is important to stick to what the authors actually claimed:

They did not find a "perfect" bite height. They didn't say, "Raise the bite by 3 degrees and you will be cured."
They did not prove cause and effect. They didn't say, "The bite caused the walking change." They only showed that when the bite changes, the walking organization changes in complex ways.
They did not test many people. This was a study of one single person. The authors admit this means we can't be sure this happens to everyone else yet.
They did not create a medical rule. They are not telling doctors how to treat patients today. They are offering a new way to think about the data.

The Takeaway

The main message is simple: Don't judge a book by its cover.

In biomechanics, a "good" walking score doesn't guarantee a "healthy" or "efficient" internal system. Two different internal states can produce the same external result. To truly understand how a body works under pressure (like a changed bite), we need to look beyond the simple numbers and understand the hidden, complex organization of the movement itself.

1. Problem Statement

In biomechanics, observable performance metrics (e.g., walking speed, step length) are frequently used as proxies for the underlying organization of biological systems. The paper argues that this assumption is flawed in adaptive systems: similar observable outputs can arise from distinct internal system states.

The study focuses on the Vertical Dimension of Occlusion (VDO) as a constraint applied to a neuromechanical system. Traditionally, VDO is treated as a static anatomical value to be restored. However, in complex rehabilitation contexts (such as for patients with Parkinson's disease), functional instability can persist despite "anatomically correct" reconstructions. The authors posit that VDO should be viewed not as a fixed target, but as a dynamic constraint interacting with sensorimotor processes. The core problem is the non-identifiability of system organization when relying solely on aggregated linear metrics.

2. Methodology

Study Design & Participant

Design: Single-case experimental design (intra-individual analysis) to track system evolution under controlled variations.
Participant: A 72-year-old male with a 13-year history of Parkinson's disease (autonomous gait, no assistive devices).
Context: Two sessions (M1 and M2) separated by 11 weeks. Between sessions, the participant underwent an uncontrolled "sophrocoaching" intervention (sensorimotor regulation training), serving as a contextual variable for temporal evolution.
Occlusal Conditions: Six conditions were tested:
- Natural Occlusion (ONL)
- Clenched Occlusion (OSL)
- Open Mouth (OBL)
- Increased VDO by 2.5° (OC2.5)
- Increased VDO by 3° (OC3)
- Increased VDO with mandibular protrusion (OC3P)

Data Acquisition

Sensors: Instrumented insoles combining pressure sensors and Inertial Measurement Units (IMUs).
Protocol: Standardized linear walking paths; only linear phases were analyzed.
Preprocessing: Outlier removal (IQR method), median replacement, and min-max normalization. No feature selection was performed prior to latent space analysis to preserve intrinsic data structure.

Analytical Framework (Four Levels)

The study proposes a multi-level analysis to decouple performance from organization:

Level 1: Observable Performance (Aggregated Metrics)
- Calculates a Global Postural Performance Score (GPPS) based on a weighted sum of spatiotemporal parameters (speed, cadence, asymmetry, step length) and postural indices (Center of Pressure, plantar capacitive index).
- Purpose: Descriptive comparison of global trends.
Level 2: Dynamical Organization (State Space)
- Models the system as a dynamical system $dX/dt = F(X, \lambda, \theta)$ , where $\lambda$ (VDO) is a constraint parameter.
- Purpose: Visualizes the evolution of trajectories in state space to assess dispersion and compactness.
Level 3: Latent Organization (Unsupervised Embedding)
- Uses UMAP (Uniform Manifold Approximation and Projection) to map high-dimensional gait data ( $R^p$ ) into a low-dimensional latent space ( $R^d$ ).
- Purpose: To reveal hidden structural patterns and groupings that linear metrics miss.
Level 4: Conceptual Extension
- A theoretical layer proposing relationships between system states in a broader configuration space.
- Note: This level is not implemented with experimental data; it is a conceptual framework for future investigation.

3. Key Results

Level 1: Aggregated Performance (GPPS)

Overlap: Different occlusal conditions produced overlapping GPPS distributions. Specifically, OC2.5 and OC3 showed nearly identical mean and median GPPS values.
Temporal Evolution: A general increase in GPPS was observed from M1 to M2, but the magnitude of improvement varied by condition, indicating condition-dependent system evolution.
Limitation: Aggregated metrics failed to distinguish between conditions that produced similar outputs but likely different internal mechanisms.

Level 2: Dynamical Organization

Trajectories in state space showed distinct organizations based on the occlusal condition.
Extreme conditions (e.g., OBL, OC3P) exhibited higher trajectory dispersion and variability.
Stable conditions (ONL, OC2.5, OC3) showed more compact, less dispersed trajectories.

Level 3: Latent Space Analysis (The Core Finding)

Dissociation: The UMAP projection revealed that OC2.5 and OC3, despite having statistically similar GPPS values, occupied distinct, non-overlapping regions in the latent space.
Interpretation: This confirms that the system organized itself differently under OC2.5 and OC3, even though the "output" (GPPS) appeared identical.
Clinical Undecidability: The study defines a "Zone of Clinical Undecidability" where aggregated metrics cannot resolve which underlying system state is present.

4. Key Contributions

Demonstration of Non-Identifiability: The study provides empirical evidence that similar observable performance does not guarantee equivalent system organization. Two different occlusal constraints can yield the same GPPS but fundamentally different latent structures.
Multi-Level Analytical Framework: It introduces a structured approach combining:
- Linear aggregated metrics (Level 1).
- Dynamical systems theory (Level 2).
- Representation learning/Unsupervised embedding (Level 3).
Reconceptualization of VDO: It shifts the clinical view of VDO from a static "optimal value" to a constraint parameter that modulates the organization of accessible system states.
Methodological Rigor: By applying unsupervised learning (UMAP) without prior feature aggregation, the study captures high-dimensional temporal structures that traditional biomechanical analysis overlooks.

5. Significance and Limitations

Significance

Clinical Implications: Clinicians cannot rely solely on scalar performance metrics (like GPPS) to determine the efficacy of occlusal interventions. Two treatments yielding the same "score" may engage different neuromuscular strategies, potentially leading to different long-term stability or adaptability.
Biomechanical Theory: Supports the view of biological systems as adaptive, constraint-driven entities where global behavior emerges from complex interactions rather than linear cause-and-effect relationships.
Future Research: Proposes a path toward analyzing "state transitions" and "configuration spaces" in biological systems, moving beyond static snapshots to dynamic organizational maps.

Limitations

Single-Case Design: Results are specific to one individual with Parkinson's disease and cannot be generalized to the broader population without further study.
Uncontrolled Variables: The "sophrocoaching" intervention between sessions was not isolated, making it difficult to attribute temporal changes solely to the occlusal constraints.
Algorithmic Dependence: The Level 3 latent structures depend on UMAP parameters (neighbors, distance metrics), meaning the specific visual groupings are method-dependent and not unique physical truths.
Exploratory Nature: The study is strictly descriptive and non-causal; it does not establish mechanistic links or predictive models.
Level 4 Conceptual: The proposed fourth level of state relationships remains theoretical and unvalidated.

Conclusion

The paper concludes that aggregated metrics are insufficient for characterizing complex biomechanical systems under constraint. To fully understand system organization, analysis must extend beyond linear performance scores to include dynamical state space and latent structural representations. This approach reveals that "good performance" can mask distinct, and potentially non-equivalent, internal system organizations.