🧠 neurology

Aggregation and analysis of 25 years of prion disease natural history extracted from published literature

This study aggregates and analyzes 25 years of prion disease literature to quantify the symptomatic course and identify critical data limitations, such as the scarcity of individual patient-level data and disease severity markers, which currently hinder the rational design of pivotal clinical trials.

Original authors: Xu, L. M., Sprague, D. A., Vallabh, S. M., Minikel, E. V.

Published 2026-08-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xu, L. M., Sprague, D. A., Vallabh, S. M., Minikel, E. V.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

Imagine a world where a mysterious thief steals your memories, your personality, and eventually your ability to move, all within a matter of months. This is the reality of prion disease, a rare and terrifying brain condition. Unlike a virus or bacteria that you can fight with antibiotics, prions are misfolded proteins that act like a corrupted computer file; when they touch healthy proteins in the brain, they force those healthy ones to fold incorrectly too, creating a chain reaction that destroys brain tissue. Because this disease is so rare and so fast-moving, scientists have struggled to understand exactly how it progresses from the first sign of trouble to the end. To design a medicine that can stop this thief, researchers need a perfect map of the journey. They need to know: How long does the disease last? When does it get bad? And most importantly, how long do patients have to get help before it's too late? Without this map, trying to test a new drug is like trying to hit a moving target in the dark.

This paper is a massive treasure hunt for that map. The authors, a team of scientists from Harvard, MIT, and the Broad Institute, decided to look at everything written about prion disease over the last 25 years. They didn't just read a few articles; they dug through 660 scientific papers, carefully selecting 245 that had enough data to be useful. Their goal was to build a giant, public database of "natural history"—which is just a fancy way of saying "what happens to patients when no one is trying to cure them." They wanted to see if the existing information was good enough to help design the next generation of clinical trials for new drugs that are currently being developed.

Here is the twist: after gathering all this data, the authors found that the map they were looking for is actually full of holes. While they successfully extracted data on 1,418 individual patients and hundreds of groups, the information was often the wrong kind. Think of it like trying to plan a road trip by only looking at a map that tells you how long the drive is from your house to the destination, but completely ignores the fact that you spent the first 70% of the time stuck in traffic trying to find the highway. The paper found that 90% of the reports only tracked patients from the moment they first felt a symptom until they died. But in the real world, patients don't walk into a doctor's office the second they feel a little fuzzy. They spend an average of 106 days in a "diagnostic odyssey," bouncing between doctors and tests, before they even get a confirmed diagnosis.

The authors discovered that this missing piece of the puzzle is a huge problem. If you only know the total time from symptom to death (which was about 152 days on average), you don't know how much time is left after a patient is actually diagnosed and enrolled in a trial. In fact, the data suggests that for many patients, the "diagnostic odyssey" eats up about 70% of their entire disease course. Furthermore, the data was often a blurry summary rather than a clear picture. Most papers only gave a "median" number (the middle value of a group) rather than tracking individual patients over time. This made it impossible to see how fast the disease was moving for a specific person or to know how severe their symptoms were at the start.

The paper also highlighted some sneaky biases. It turns out that the stories we read in medical journals aren't a perfect reflection of reality. For instance, the data was skewed toward patients with genetic forms of the disease or those who were very sick, while the most common type (sporadic) was under-reported. Additionally, the way patients were found changed the results: patients found by looking back at old hospital records seemed to die faster than those found in new, forward-looking studies. This suggests that the "average" survival time reported in the past might be misleading, potentially making a new drug look like it's working when it's not, or vice versa.

Ultimately, this paper is a call to action. The authors conclude that while they have assembled the largest public dataset of its kind, the current literature is simply not good enough to design the critical clinical trials needed to save lives. They argue that scientists and doctors need to start reporting different kinds of data: specifically, how long it takes from the moment of diagnosis to the end, and how patients' mental and physical abilities change day by day. Until the medical community starts sharing these detailed, individual stories rather than just rough averages, designing a successful treatment for prion disease will remain a game of guesswork. The map exists, but it needs to be redrawn with much finer detail before we can navigate our way to a cure.

Problem
Prion disease is a rare, uniformly fatal neurodegenerative condition for which new therapeutic candidates, specifically those designed to lower prion protein (PrP), are entering clinical trials. A critical prerequisite for designing well-powered pivotal trials is a robust understanding of the disease's natural history—the trajectory from onset to death under standard care without effective intervention. While the FDA and regulators increasingly accept natural history data to define clinical endpoints, refine inclusion criteria, and even serve as external control arms (ECA), the existing literature lacks the granularity required for these purposes. Previous trials have struggled with heterogeneous patient populations and endpoints that do not reflect the moment of potential trial enrollment (e.g., diagnosis) rather than symptom onset. Furthermore, past open-label studies suggesting therapeutic benefit failed to replicate in rigorous placebo-controlled designs, partly due to inadequate historical controls. The core problem addressed is the insufficiency of currently published data to quantitatively model trial scenarios, specifically due to a lack of longitudinal individual patient data (PLD), a reliance on aggregated statistics, and a focus on symptom-onset-to-death intervals rather than clinically relevant milestones like diagnosis-to-death.

Methodology
The authors conducted a systematic literature review following PRISMA guidelines, searching PubMed for articles published between 2000 and 2024. The search string targeted prion disease natural history, duration, survival, and progression. Inclusion criteria required full-text English articles reporting quantitative human data for at least five patients, including at least one time-to-milestone interval.

Data Extraction: Two reviewers manually extracted data into two categories: aggregate cohort statistics (groups of $n \ge 2$ ) and individual patient-level data (PLD). Extracted variables included demographics, ascertainment methods, histopathologic subtypes, PRNP variants, cognitive test scores, and time intervals between milestones (symptom onset, clinical presentation, diagnosis, diagnostic tests, akinetic mutism, and death).
Processing: Data were harmonized into days. Duplicate patients across studies were identified and removed from the PLD set using a custom R script based on matching survival duration, sex, age of onset, region, and subtype. Aggregate cohorts could not be deduplicated.
Analysis: The authors computed summary statistics (weighted medians, confidence intervals) for cohorts. They generated Kaplan-Meier survival curves stratified by histologic subtype and performed univariate and multivariate Cox proportional hazards analyses on the PLD to assess the effects of covariates (age, sex, subtype) on survival. Meta-regression was used to assess heterogeneity in aggregate cohort medians.

Key Contributions

Public Dataset: The study assembled and released the largest public dataset characterizing the symptomatic natural history of prion disease, comprising 418 aggregate cohort medians and 1,400 rows of deduplicated individual patient data derived from 245 publications.
Quantification of Literature Gaps: The authors systematically quantified the deficiencies in published data, specifically highlighting the scarcity of data starting from clinical milestones (diagnosis/presentation) rather than symptom onset, and the near-total absence of longitudinal cognitive or functional scores.
Bias Characterization: The study identified and quantified significant publication and ascertainment biases, demonstrating how different methods of patient recruitment (surveillance vs. prospective studies) skew survival estimates.

Results

Data Composition: From 660 screened papers, 245 were included. The dataset includes 682 aggregate cohorts and 1,418 individual patients. There is a marked over-representation of genetic and acquired cases in the literature compared to the ~85% sporadic incidence in the general population.
Milestone Reporting: 90% of aggregate cohorts and 91% of individual patients reported symptom-to-death intervals. In contrast, only 7% of cohorts and 11% of individuals reported intervals from a clinical presentation milestone (e.g., diagnosis) to death. The weighted median symptom-to-death interval was 152 days, while the weighted median presentation-to-death interval was 46 days, suggesting approximately 70% of the disease course is spent in a "diagnostic odyssey."
Survival Variability: Survival times varied significantly based on ascertainment. Retrospective cohorts had a median survival of 122 days, whereas prospective cohorts had a median of 207 days. Even within the same histopathologic subtype (e.g., MM2), median survival times varied up to 3.2-fold across different cohorts.
Endpoint Definitions: When both endpoints were reported, the time to akinetic mutism occurred 73% sooner than death (e.g., 91 days vs. 335 days for MM1 subtype). This indicates that life-extending measures, particularly common in Japan, significantly inflate survival time if death is the sole endpoint.
Covariates: Cox models confirmed that older age and male sex are associated with shorter survival. Significant survival differences were observed between subtypes (e.g., MM1 vs. MM2) and genetic variants (e.g., P102L vs. E200K), even after controlling for demographics.
Lack of Severity Metrics: Cognitive tests were reported in only 11% of aggregate cohorts and 6% of individual patients. Longitudinal data (multiple timepoints) were virtually non-existent.

Significance and Claims
The paper claims that while the existing literature provides a broad overview of prion disease survival, it is inadequate for rational clinical trial design. The authors argue that the current data are too aggregated, cross-sectional, and misaligned with the clinical reality of trial enrollment (which occurs post-diagnosis, not at symptom onset). Specifically, the lack of data on disease severity at diagnosis and the rate of progression after diagnosis prevents sponsors from modeling trial scenarios, such as estimating the percentage of patients who will remain above a specific functional cutoff (e.g., MRC-PDRS) for a required duration.

The authors conclude that to enable the development of effective therapies, the field must shift toward:

The public deposition of de-identified individual patient data.
Reporting clinically relevant intervals (diagnosis-to-death) rather than just symptom-to-death.
Routine collection and reporting of longitudinal cognitive and functional scores.

The study positions itself not as a solution to the trial design problem, but as a diagnostic tool that reveals the insufficiency of current data sources, thereby motivating a change in how clinical data is collected and reported to support future pivotal trials.

More like this