🧠 neurology

Time's up: Using data-driven phenotype-severity metrics not time to map progression in the dementias

This study proposes a data-driven, transdiagnostic phenotype-severity metric derived from neuropsychological scores that outperforms traditional time-based measures in accurately mapping dementia progression and distinguishing disease stages from clinical heterogeneity.

Original authors: Smith, V., Schumacher, R., Ramanan, S., Bouzigues, A., Russell, L. L., Foster, P. H., Ferry-Bolder, E., van Swieten, J. C., Jiskoot, L. C., Seelaar, H., Sanchez-Valle, R., Laforce, R., Graff, C., Gali

Published 2026-01-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Smith, V., Schumacher, R., Ramanan, S., Bouzigues, A., Russell, L. L., Foster, P. H., Ferry-Bolder, E., van Swieten, J. C., Jiskoot, L. C., Seelaar, H., Sanchez-Valle, R., Laforce, R., Graff, C., Galimberti, D., Vandenberghe, R., de Mendonca, A., di Fede, G., Santana, I., Gerhard, A., Levin, J., Nacmias, B., Otto, M., Bertoux, M., Lebouvier, T., Ducharme, S., Butler, C. R., Le Ber, I., Finger, E., Tartaglia, M. C., Masellis, M., Synofzik, M., Moreno, F., Borroni, B., Rohrer, J. D., Rowe, J. B., Lambon Ralph, M. A.

Original paper licensed under CC BY 4.0 (https://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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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 Problem: The "Stopwatch" Doesn't Work

Imagine trying to measure how far a runner has traveled in a race. Usually, you might look at their stopwatch (time) to guess their progress. You might think, "If they've been running for 10 years, they must be very tired and far along the track."

But in dementia research, the "stopwatch" is broken.

The Race Start is Blurry: It's hard to know exactly when the disease started. Did it begin when the first memory slipped? Or when the family noticed? Or when the doctor finally diagnosed it?
The Watchers Vary: Some people notice small changes immediately; others ignore them for years.
The Pace Varies: Some runners sprint to the finish line; others jog slowly.

Because of this, using "time since diagnosis" to measure how sick a patient is creates a lot of confusion. It's like trying to line up runners from different races just by looking at their watches; they might all have been running for 5 years, but one is at the starting line and the other is finishing.

The Solution: A "Fitness Score" Instead of a Stopwatch

The researchers asked: What if we stop looking at the clock and just measure the runner's actual fitness?

They analyzed data from a large group of people with genetic frontotemporal dementia (a type of brain disease). Instead of asking "How long have you been sick?", they looked at how well the patients performed on a battery of mental tests (memory, language, problem-solving).

They used a mathematical tool called Principal Component Analysis (PCA). Think of this as a smart blender that takes all the different test scores and mixes them into one single, smooth "Disease Severity Smoothie."

The Result: This "smoothie" created a single number that accurately represented how severe a person's symptoms were, regardless of how long they had been sick or what their specific genetic mutation was.
The Proof: This new number lined up perfectly with how impaired the patients were in their daily lives (measured by standard doctor ratings), proving it was a true measure of the disease's impact.

The Big Discovery: Time is the Result, Not the Cause

Once they had this accurate "Severity Score," they flipped the script. Instead of using time to guess severity, they used severity to guess time.

They treated time as a dependent variable (the thing that changes based on the disease) rather than the ruler. This allowed them to see the "speed" of the disease.

The Accelerator: They found that the disease doesn't move at a constant speed. It's like a car going downhill: the steeper the hill (the more severe the disease), the faster the car goes. As patients get sicker, their decline accelerates.
The Genetic Differences: They found that different genetic groups drive different cars:
- The GRN group: These patients had the "fastest car." They progressed through the severity levels more quickly than the others.
- The C9orf72 group: These patients had a "slower car." They took much longer to travel the same distance of severity.
- The MAPT group: They were somewhere in the middle.

This explains why some people seem to stay sick for a very long time while others decline rapidly. It's not random; it's about the specific "engine" (genetics) driving the progression.

The Shortcut: The "Three-Test" Battery

Doing a full battery of mental tests takes a long time and is tiring for patients. The researchers wanted to know: Can we just do a few quick tests and still get the same accurate "Severity Score"?

They found that three specific tests were enough to capture 93% of the information needed to measure severity:

Digit Symbol: A quick pattern-matching test.
Verbal Fluency (Letters): Naming as many words as possible starting with a specific letter.
Trail Making Test (Part B): Connecting numbers and letters in a specific order.

These three tests take only about 12–15 minutes total. Using them, doctors and researchers can get a highly accurate "fitness score" for the disease without exhausting the patient.

Finding the Early Warning Signs

Finally, they looked at which tests were the first to show trouble.

The "Canary in the Coal Mine": The Trail Making Test (Part B) was the first to drop below normal levels, even in people who were still considered "asymptomatic" or only mildly affected.
The MAPT Clue: For people with the MAPT gene mutation, the Boston Naming Test (naming pictures) was the first to show decline, which matches what we know about that specific genetic group.

Summary

This paper argues that we need to stop using "time since diagnosis" as a ruler for dementia because it's inaccurate. Instead, we should use a data-driven "Severity Score" based on actual test performance.

This new approach:

Gives a clear, objective picture of how sick a patient is.
Reveals that different genetic groups progress at different speeds.
Shows that the disease gets faster as it gets worse.
Allows researchers to use a short, efficient set of tests to track these changes accurately.

By using this method, we can finally line up patients correctly for clinical trials and understand the true nature of how these diseases progress.

Technical Summary: Time's up: Using data-driven phenotype-severity metrics not time to map progression in the dementias

Problem Statement
Current research and clinical trials in neurodegenerative diseases frequently rely on temporal proxies (e.g., time since symptom onset or diagnosis) to track disease severity and progression. The authors argue these metrics are fundamentally flawed due to substantial variance and bias introduced by:

Variability in symptom awareness among patients and caregivers.
Differences in the timing of clinical presentation and diagnostic delays.
Heterogeneous rates of disease progression.
Differential sensitivity and emergence rates of biomarkers.

These limitations make it difficult to accurately map disease progression, align individuals at comparable stages, or distinguish between qualitatively different subtypes versus quantitatively varying severity levels of a single disease trajectory. Consequently, temporal measures often fail to serve as reliable independent variables for controlling severity in clinical trials or for understanding natural history.

Methodology
The study utilized data from the Genetic Frontotemporal Dementia Initiative (GENFI 1 and 2), comprising 686 individuals (symptomatic carriers of C9orf72, GRN, or MAPT mutations, and converters from at-risk to symptomatic) across 1,678 visits.

Data-Driven Severity Metric Derivation:
- The authors applied unrotated Principal Component Analysis (PCA) to a comprehensive battery of neuropsychological assessments (including Digit Span, Digit Symbol, Trail Making Tests, Boston Naming, Verbal Fluency, etc.).
- Missing data were imputed using a PCA-based iterative method.
- The analysis focused on symptomatic individuals and those converting from at-risk to symptomatic.
- A single principal component (eigenvalue > 1) was extracted to serve as a global, transdiagnostic "phenotype-severity" score.
Validation and Stability:
- The severity score was validated against the Clinical Dementia Rating scale with the National Alzheimer Coordinating Centre FTLD score (CDR®-FTLD) and the Cambridge Behavioural Inventory-Revised (CBI-R) using ordinal regression.
- Stability was tested by comparing results from the full dataset against subsets (baseline only, first/middle/last visits) and by re-running PCA with an expanded battery (GENFI 2) that included additional behavioral and functional measures (CBI-R, FRS).
Re-framing Time:
- Instead of using time as a proxy for severity, the study treated time as an observed, dependent variable.
- Linear mixed-effects models and quadratic regression were used to analyze the "velocity" (change in severity per unit of time) and "acceleration" of progression across genetic groups (C9orf72, GRN, MAPT).
- Cumulative time curves were generated to estimate the duration required to traverse specific severity intervals for each genetic group.
Test Selection:
- Stepwise forward regression was employed to identify a time-efficient sub-battery of neuropsychological tests capable of explaining the majority of the variance in the global severity score.

Key Results

Unified Severity Metric: A single unrotated PCA component emerged, explaining 65% of the variance in neuropsychological scores (KMO = 0.92). All tests loaded highly on this factor, creating a unified metric of global cognitive severity that was consistent across genetic and diagnostic groups (bvFTD, PPA, ALS).
Validation: The severity score showed a strong relationship with functional impairment (CDR®-FTLD). While the severity score alone predicted CDR®-FTLD, a combined model including behavioral measures (CBI-R) provided a better fit, indicating that CDR®-FTLD reflects both cognitive and behavioral severity.
Failure of Temporal Proxies: The data-driven severity score showed almost no correlation with "years from symptom onset" ( $R^2 \le 0.005$ ). Similarly, "visit number" introduced attrition bias, with mean severity scores appearing to decrease over time due to the dropout of severe cases.
Progression Dynamics:
- Progression accelerates as severity increases (significant positive quadratic effect).
- Genetic Differences: GRN mutation carriers exhibited the fastest baseline velocity (fastest progression), followed by MAPT, with C9orf72 carriers showing the slowest progression.
- Time Estimates: GRN carriers were estimated to traverse the severity spectrum (-1.2 to 2) in 6.8 years, compared to 11.9 years for MAPT and 19.7 years for C9orf72.
Efficient Sub-battery: Three tests—Digit Symbol, Verbal Fluency (letters), and Trail Making Test-Part B—explained approximately 93% of the unique variance in the cognitive severity score, offering a time-efficient alternative to the full battery.
Early Detection: The Trail Making Test-Part B and Boston Naming Test (specifically in MAPT carriers) showed sensitivity to early decline, with some individuals scoring below control cut-offs despite having low global severity scores.

Significance and Claims
The paper claims to provide an objective, precise, and transdiagnostic measure of disease severity that is independent of time. By shifting the analytical framework to treat time as a dependent variable, the authors demonstrate that:

Temporal metrics are poor proxies: They introduce significant noise and bias, failing to capture the true state of disease progression.
Data-driven severity enables better stratification: The proposed metric allows for the alignment of patients based on actual phenotypic severity rather than arbitrary time points, which is crucial for distinguishing disease subtypes from severity stages.
Improved Clinical Trial Design: This approach offers a more accurate outcome measure for clinical trials, particularly for disease-modifying therapies that may have optimal windows in the disease course. It allows for better control of severity-related effects and more precise estimation of progression rates across subtypes.
Efficiency: The identification of a three-test sub-battery suggests that accurate severity tracking can be achieved with significantly reduced assessment time, facilitating large-scale studies and trials.

The authors conclude that this approach may shed new light on clinical heterogeneity, helping to determine whether observed differences reflect distinct subtypes or simply variations in disease stage. They acknowledge that the metric is currently specific to the FTD-related genetic sample and that future work is needed to test its applicability across other neurodegenerative etiologies and to explore potential multidimensional aspects of severity.