The Case Against the 'S': Is Functional Neurological Disorder(s) One Condition or Many?

By applying Bayesian model comparison to symptom data from 697 patients, this study provides empirical evidence that Functional Neurological Disorder (FND) is best characterized as a single pathophysiological entity with a heterogeneous phenotype rather than a collection of multiple discrete conditions.

The Gist

Imagine you walk into a doctor's office and say, "I have a headache." That's a symptom. But is "headache" one single disease, or is it a label we use for five completely different things (like a migraine, a sinus infection, a tension headache, a tumor, or dehydration)?

This is the big question researchers asked about Functional Neurological Disorder (FND). FND is a condition where people have real, physical symptoms (like paralysis, tremors, or seizures) that aren't caused by damage to the nerves or brain structure, but rather by a "glitch" in how the brain sends and receives signals.

For a long time, doctors and scientists have argued: Is FND one single "glitch" that just looks different in everyone? Or is it actually a bunch of totally different "glitches" that we've accidentally lumped together under one name?

This paper tries to answer that question using math and data, rather than just opinions. Here is the breakdown in simple terms:

The Two Competing Theories

The authors set up a statistical "race" between two ideas:

1. The "One Glitch" Theory (The Smooth Slope):
   Imagine a giant, smooth hill. Everyone with FND is standing somewhere on this hill. Some are at the bottom, some at the top, and some in the middle. The "height" of the hill represents how severe the symptoms are, and the "direction" you walk represents which symptoms you have.

   • The Analogy: Think of color. Red, orange, yellow, green, blue, and violet are all just different points on a single, continuous rainbow. You can't draw a hard line where "orange" stops and "yellow" begins; it's just a smooth transition. The researchers thought FND might be like a rainbow: one single condition that varies smoothly from person to person.

2. The "Many Glitches" Theory (The Separate Islands):
   Imagine an archipelago of islands. Each island represents a totally different disease. People on Island A have one set of symptoms, and people on Island B have a completely different set. There is deep water (a clear gap) between them.

   • The Analogy: Think of fruits. An apple is an apple, and a banana is a banana. They are distinct categories. If FND is like this, then "FND Type A" and "FND Type B" are actually different diseases that just happen to look similar on the surface.

The Experiment: Counting the Clusters

The researchers took data from 697 people with FND who reported their symptoms (like weakness, speech issues, or vision problems). They used advanced computer models to see which theory fit the data better.

• They tried to fit the data into the "Smooth Hill" model (Latent Trait models).
• They tried to fit the data into the "Separate Islands" model (Latent Class models).

The Results: The Islands Were Actually Just Hills

The computer analysis gave a clear winner: The "Smooth Hill" model fit the data much better.

Here is what they found when they looked at the "Separate Islands" model:

• No Clear Water: When they tried to draw lines between the "islands," the lines were fuzzy. People didn't clearly belong to one group or another; they were sort of in between.
• The "Effective" Number: Even though they tried to sort people into 5 different groups, the math showed that these groups were so mixed up that they effectively acted like 1.38 groups. It's like trying to sort a pile of sand into buckets; no matter how many buckets you use, the sand just flows between them.
• The Conclusion: The "islands" weren't real. They were just an illusion created because we tried to force a smooth, continuous spectrum into rigid boxes.

What Does This Mean for Patients?

The study suggests that FND is likely one single condition with a very wide variety of symptoms, rather than a collection of different diseases.

• The "Common Pathway" Idea: The authors suggest that even if different things start the problem (like stress, trauma, or genetics), they all seem to hit the same "switch" in the brain. Once that switch is flipped, the brain produces the symptoms. It's like different keys (stress, injury, virus) all opening the same door (the FND symptoms).
• Why This Matters:
  • Stop Sorting by Symptoms: Trying to create strict sub-groups based on symptoms (e.g., "The Tremor Group" vs. "The Paralysis Group") might be a waste of time because those groups overlap too much.
  • Focus on the Spectrum: Instead of asking "Which type of FND do you have?", doctors and researchers should look at how the condition varies along a spectrum.
  • Better Treatment: If it's one condition, treatments might work better if they target the underlying "glitch" mechanism common to everyone, rather than trying to find a specific cure for every tiny variation.

The Bottom Line

Think of FND not as a box of different colored marbles, but as a single, long, winding road. Some people are walking fast, some slow; some are carrying heavy bags, some are light. But they are all walking on the same road.

The study tells us that to understand and treat FND, we need to stop trying to build walls between different types of patients and start understanding the continuous journey they are all on.

Technical Summary

1. Problem Statement

Functional Neurological Disorder (FND) is a prevalent yet poorly understood neurological condition. A fundamental debate exists regarding its nosological structure:

• The "One" Hypothesis: FND is a single disorder with a heterogeneous phenotype (variable expression of symptoms).
• The "Many" Hypothesis: FND is a collection of distinct, clinically overlapping conditions erroneously grouped under a single diagnostic label.

Current debates rely heavily on conceptual models and clinical intuition. Without biomarkers, it is unclear whether observed symptom heterogeneity arises from a single pathophysiological process with variable expression or from multiple distinct pathologies converging on similar clinical presentations. The authors frame this as a model comparison problem to determine which structural assumption best fits empirical symptom data.

2. Methodology

The study utilized a large dataset from the FND Research Connect database, comprising symptom reports from 697 individuals with FND. Participants reported the presence/absence of 19 specific FND symptoms experienced in the preceding two months.

The authors employed a statistical model comparison approach using Bayesian methods:

• Competing Models:

  1. Latent Trait Models (Continuous): Represent FND as a single entity where symptoms vary along continuous latent dimensions (similar to Item Response Theory). These models assume a single underlying disease process with dimensional variation.
  2. Latent Class Analysis (LCA) Models (Discrete): Represent FND as multiple discrete subgroups (classes), where each class generates a distinct pattern of symptoms. Models were tested with class numbers ranging from 1 to 8.

• Statistical Implementation:

  • Software: Python (using Stepmix for LCA and PyMC for latent trait models).
  • Model Selection: Models were compared using patient-level K-fold cross-validated mean log predictive density (MLPD). This metric evaluates out-of-sample predictive performance.
  • Secondary Analysis: To determine if LCA classes represented true discrete subgroups or merely artifacts of discretizing continuous data, the authors analyzed:
    • Effective Number of Classes ($K_{eff}$): Based on mean perplexity.
    • Boundary Mass: The proportion of participants with low posterior probability (<0.8) for their assigned class.
    • Low-dimensional embeddings: Visualizing symptom similarity (Jaccard distance) to check for gaps between clusters.
  • Interpretation: The best-fitting latent trait model (4 dimensions) underwent Varimax rotation to identify interpretable symptom clusters associated with each dimension.

3. Key Contributions

• Empirical Resolution of a Nosological Debate: The study moves beyond theoretical arguments to provide empirical evidence regarding the structural validity of FND as a single vs. multiple entities.
• Methodological Rigor: It applies rigorous Bayesian model comparison and cross-validation techniques, which are rarely used in FND research, to distinguish between continuous dimensional models and discrete categorical models.
• Reframing Heterogeneity: The study demonstrates that the apparent heterogeneity of FND symptoms is better explained by continuous variation along latent dimensions rather than by distinct categorical subtypes.

4. Results

• Model Fit Comparison:

  • Latent Trait Models: All latent trait models with more than one dimension fit the data significantly better than the best-fitting LCA model.
  • Predictive Gain: The best-fitting continuous model showed a 0.12 difference in log predictive density over the best discrete model, corresponding to an approximate 13% increase in predictive probability.
  • Optimal Dimensionality: The best-fitting latent trait model utilized 4 dimensions.

• Analysis of Latent Class Models:

  • Secondary analysis of the best-fitting 5-class LCA model revealed it did not represent true discrete subgroups:
    • Effective Number of Classes ($K_{eff}$): Only 1.38, indicating patients effectively occupied a continuum rather than distinct categories.
    • Boundary Mass: 22.4% of participants had high uncertainty regarding their class membership (posterior probability < 0.8).
    • Visual Embedding: 2D embeddings of symptom profiles showed substantial overlap and smooth variation with no clear gaps between categories.

• Symptom Dimensions (Rotated Factors):
  The four latent dimensions identified were:

  1. Weakness, gait disorder, and somatosensory symptoms.
  2. Speech disturbance.
  3. Depersonalization/derealization and dystonia.
  4. Tinnitus, elimination problems, and visual symptoms.

5. Significance and Implications

• Pathophysiological Inference: The results support the hypothesis that FND is a single disease entity with a heterogeneous, dimensional phenotype. Even if multiple upstream causal processes exist, they likely converge on a single common pathophysiological pathway that dominates the clinical presentation.
• Clinical and Research Impact:
  • Against Categorical Subtyping: The study argues against stratifying patients into distinct "types" of FND based solely on symptom clusters, as these clusters appear to be artifacts of continuous variation.
  • Future Directions: Research and clinical stratification should focus on dimensional variation, risk factors, causal mechanisms, longitudinal trajectories, and treatment response rather than searching for discrete symptom-defined subgroups.
• Limitations & Robustness: While the data were self-reported and binary (potentially obscuring severity), the authors argue that such limitations typically bias analysis toward finding categorical subgroups. The fact that continuous models still outperformed discrete models strengthens the conclusion that the underlying structure is dimensional.

Conclusion: The paper provides strong empirical evidence that Functional Neurological Disorder is best understood as a single, heterogeneous condition defined by continuous symptom dimensions, rather than a collection of distinct disorders.