A meaningful prediction of functional decline in amyotrophic lateral sclerosis based on multi-event survival analysis
Christian Marius Lillelund* · 2025
Read A meaningful prediction of functional decline in amyotrophic lateral sclerosis based on multi-event survival analysis by Christian Marius Lillelund* (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-02076.txt.
Introduction
Christian Marius Lillelund and co-authors wrote a 2025 paper on forecasting decline in ALS, a disease that wastes the motor nerves. Patients lose skills such as speech, swallowing, hand use, walking, and breathing at varied rates. Knowing when each skill may fail helps time care and aids.
The team framed the task as a multi event survival problem. Five models, one per skill, were trained on a large trial data set with 3,220 patients. Each model gives a personal curve for when a skill is likely to fall to a low level within 500 days. The models beat a standard baseline curve. The method also lets doctors test what if cases for a single patient.
Catalog record
| CREATOR | Christian Marius Lillelund* |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-02076
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A meaningful prediction of functional decline in amyotrophic lateral sclerosis based on multi-event survival analysis Christian Marius Lillelund* Abstract Abstract Amyotrophic lateral sclerosis (ALS) is a degenerative disorder of motor neurons that causes progressive paralysis in patients. Current treatment options aim to prolong survival and improve quality of life; however, due to the heterogeneity of the disease, it is often difficult to determine the optimal time for potential therapies or medical interventions. In this study, we propose a novel method to predict the time until a patient with ALS experiences significant functional impairment (ALSFRS-R \leq 2) with respect to five common functions: speaking, swallowing, handwriting, walking and breathing. We formulate this task as a multi-event survival problem and validate our approach in the PRO-ACT dataset by training five covariate-based survival models to estimate the probability of an event over a 500-day period after a baseline visit. We then predict five event-specific individual survival distributions (ISDs) for each patient, each providing an interpretable and meaningful estimate of when that event will likely take place in the future. The results show that covariate-based models are superior to the Kaplan-Meier estimator at predicting time-to-event outcomes. Additionally, our method enables practitioners to make individual counterfactual predictions, where certain features (covariates) can be changed to see their effect on the predicted outcome. In this regard, we find that Riluzole has little to no impact on predicted functional decline. However, for patients with bulbar-onset ALS, our method predicts considerably shorter counterfactual time-to-event estimates for tasks related to speech and swallowing compared to limb-onset ALS. The proposed method can be applied to current clinical examination data to assess the risk of functional decline and thus allow more personalized treatment planning. Introduction Amyotrophic lateral sclerosis (ALS) is a neurological disease that causes a gradual loss of upper and lower motor neurons in the central nervous system. Unfortunately, most patients will die within 2-5 years after onset [ 1 ] , primarily due to complications from respiratory failure [ 2 ] . The underlying causes of the disease are not yet understood, and to this day, current treatments aim to prolong survival and improve quality of life. Moreover, due to its heterogeneous and unpredictable nature, and the high variability in its progression rate and clinical phenotype [ 5 ] , it is often difficult to determine the appropriate treatment. This also makes it difficult to determine the optimal time for such medical interventions, e.g. , noninvasive ventilation, or if a specific treatment can effectively slow the disease progression. Therefore, there is a significant need for tools that can predict disease progression, facilitate treatment plans, and improve patient care. The ALS Functional Rating Scale-Revised (ALSFRS-R) [ 6 ] is the most widely-used questionnaire to evaluate the progression of ALS (see Table 1 ). It contains 12 questions to assess functional capacity in patients, with a focus on bulbar, motor, and respiratory functions – e.g. , swallowing, walking, and breathing, respectively. Each response is scored between 0 and 4, where 0 represents complete inability with regard to that function, and 4 represents normal function.Based on this questionnaire, an interesting task is to predict the degree of functional decline a patient experiences during the course of the disease, rather than simply when the patient is likely to die [ 7 , 8 ] . Many previous projects have attempted to predict the future ALSFRS-R score based on patient features (covariates) [ 12 , 9 , 10 , 11 , 3 ] . Ong et al. [ 12 ] proposed a binary model to predict the functional decline class ( i.e. , whether ALS progresses slowly or rapidly in a specific patient) after baseline (trial entry), based on several patient characteristics. Their model was able to predict slow or rapid decline, but did not predict when functional decline would occur over time. Similarly, Jabbar et al. [ 11 ] proposed a method to predict disease progression as fast or slow, but as in [ 12 ] , their model could only give a yes/no answer to whether a patient would experience functional decline over time. Amaral et al. [ 10 ] applied several clustering algorithms to clinical patient records, which revealed four separate groups regarding disease progression. Their method could assign newly-diagnosed patients to some progression group, but did not predict when a patient would experience functional decline based on their covariates. Finally, Vieira et al. [ 9 ] trained two machine learning models to predict a patient’s future ALSFRS-R score: a voice model based on speech recordings and an accelerometer model based on accelerometer measurements. These models showed promise in predictive performance, but similar to [ 10 ] and others, they simply predicted what the ALSFRS-R score would be for a patient in the future, not the time until functional decline would occur. In addition, disease progression is not necessarily constant or linear in ALS, which the authors assumed, but can vary significantly over time and between individuals [ 4 ] . Based on these shortcomings in the literature, we propose a novel method that can predict when an ALS patient will experience each of five common types of functional decline, based on the ALSFRS-R protocol (see Table 1 ): significant difficulty ( \leq 2) in each of Speaking , Swallowing , Handwriting , Walking and Dyspnea . The events are distinct but not mutually exclusive and likely share some information in their covariates; it may be that Mr. Smith with limb-onset ALS will experience difficulty walking and writing earlier than Mr. Johnson, who has bulbar-onset ALS. On the other hand, Mr. Johnson may have difficulty speaking and swallowing solid foods before Mr. Smith. As a newly-diagnosed patient with ALS, our approach will help answer important questions, such as: • How much longer will I be able to speak clearly enough to be understood? • How much longer will I be able to eat most solid foods without difficulty? • How much longer will I be able to write a shopping list that my wife can read? • How much longer will I be able to walk at a normal pace and keep my balance? • How much longer will I be able to breathe normally during daily activities? In practice, we formulate the problem as a multi-event survival problem: given an instance ( e.g. , a description of Mr. Smith at a “baseline” point in time), our method predicts the time until said Mr. Smith will experience each of the events listed above. We present each prediction as individual survival (event) distributions (ISDs) [ 13 ] , which provide the probability of each event at each future time point, after the baseline visit. These ISDs were recently used to predict mortality in patients with ALS [ 7 ] . We describe each patient by the values of a set of covariates such as age, site of onset, forced vital capacity (FVC), and others, which are recorded in clinical records obtained at or before the patient’s current visit (baseline). To our knowledge, no existing method can predict the time of distinct individual events in ALS that explicitly provide the time. Fig. 1 shows an outline of the proposed method. The source code