Identifying and predicting amyotrophic lateral sclerosis clinical subgroups: a population-based machine-learning study

Abstract

Background: Amyotrophic lateral sclerosis (ALS) is known to represent a collection of overlapping syndromes. Various classification systems based on empirical observations have been proposed, but it is unclear to what extent they reflect ALS population substructures. We aimed to use machine-learning techniques to identify the number and nature of ALS subtypes to obtain a better understanding of this heterogeneity, enhance our understanding of the disease, and improve clinical care.

Methods: In this retrospective study, we applied unsupervised Uniform Manifold Approximation and Projection [UMAP]) modelling, semi-supervised (neural network UMAP) modelling, and supervised (ensemble learning based on LightGBM) modelling to a population-based discovery cohort of patients who were diagnosed with ALS while living in the Piedmont and Valle d'Aosta regions of Italy, for whom detailed clinical data, such as age at symptom onset, were available. We excluded patients with missing Revised ALS Functional Rating Scale (ALSFRS-R) feature values from the unsupervised and semi-supervised steps. We replicated our findings in an independent population-based cohort of patients who were diagnosed with ALS while living in the Emilia Romagna region of Italy.

Findings: Between Jan 1, 1995, and Dec 31, 2015, 2858 patients were entered in the discovery cohort. After excluding 497 (17%) patients with missing ALSFRS-R feature values, data for 42 clinical features across 2361 (83%) patients were available for the unsupervised and semi-supervised analysis. We found that semi-supervised machine learning produced the optimum clustering of the patients with ALS. These clusters roughly corresponded to the six clinical subtypes defined by the Chiò classification system (ie, bulbar, respiratory, flail arm, classical, pyramidal, and flail leg ALS). Between Jan 1, 2009, and March 1, 2018, 1097 patients were entered in the replication cohort. After excluding 108 (10%) patients with missing ALSFRS-R feature values, data for 42 clinical features across 989 patients were available for the unsupervised and semi-supervised analysis. All 1097 patients were included in the supervised analysis. The same clusters were identified in the replication cohort. By contrast, other ALS classification schemes, such as the El Escorial categories, Milano-Torino clinical staging, and King's clinical stages, did not adequately label the clusters. Supervised learning identified 11 clinical parameters that predicted ALS clinical subtypes with high accuracy (area under the curve 0·982 [95% CI 0·980-0·983]).

Interpretation: Our data-driven study provides insight into the ALS population substructure and confirms that the Chiò classification system successfully identifies ALS subtypes. Additional validation is required to determine the accuracy and clinical use of these algorithms in assigning clinical subtypes. Nevertheless, our algorithms offer a broad insight into the clinical heterogeneity of ALS and help to determine the actual subtypes of disease that exist within this fatal neurodegenerative syndrome. The systematic identification of ALS subtypes will improve clinical care and clinical trial design.

Funding: US National Institute on Aging, US National Institutes of Health, Italian Ministry of Health, European Commission, University of Torino Rita Levi Montalcini Department of Neurosciences, Emilia Romagna Regional Health Authority, and Italian Ministry of Education, University, and Research.

Translations: For the Italian and German translations of the abstract see Supplementary Materials section.

Figure 1.. Workflow followed in this study.

Unsupervised and semi-supervised machine learning was applied to clinical data collected from two population-based ALS registries (n = 2,858 cases and 1,097 cases) to identify clinical subtypes. Supervised machine learning was used to predict subtypes based on clinical parameters, and a web-based tool was built for clinical researchers to apply to their own data.

Figure 2.. The ALS subtypes identified by machine learning in the discovery and replication cohorts.

The top row (A) shows the three-dimensional projections of the discovery cohort (n = 2,361) defined by the semi-supervised machine learning algorithm consisting of a UMAP algorithm applied to the output of a five-layer neural network. The same three-dimensional projections (left panel = 100 degrees azimuthal rotation, center panel = 135 degrees, and right panel = 170 degrees) of the replication cohort (n = 989) are shown in the bottom row (B). The projections are symbolic representations of ALS subtypes. Each patient (dot) was color-coded after machine learning cluster generation according to the Chiò classification system. Interactive three-dimensional graphs are available on https://share.streamlit.io/anant-dadu/machinelearningforals/main.

Figure 3.. Different classification schema applied to the semi-supervised 3D projection of the ALS discovery cohort (n = 2,361).

(A) The El Escorial classification system assigns patients to definite (def.), probable (prob.), probable - laboratory supported (prob. - lab.), possible (poss.), and suspected (susp.) categories based on their disability. (B) Patients with a family history of ALS are represented by red dots, and blue dots show patients with sporadic disease. (C) Patients carrying the pathogenic repeat expansion are represented by red dots. (D) The MITOS classification system assigns patients to clinical stages 0 to 4 based on their disability. (E) The ALSFRS-R score rates the severity of disability ranging from 0 to 48 (no disability). (F) The King’s clinical staging system classifies patients into four stages according to their disability level.

Figure 4.. Clinical parameters used in the supervised machine learning model to predict ALS clinical subtype.

(A) Graphical representation of the overlap between the eleven parameters with the most significant impact on the classification model. The dark circles in the dot plot indicate the parameters that are part of an intersection, and the vertical bar plot reports the number of patients with that parameter combination. The horizontal bar plot reports the set sizes. Analysis was confined to 699 ALS patients with no missing data. (B) Distribution of the parameters in each patient. On average, a patient had five of these clinical features. (C - E) The distribution of the age at onset, weight at diagnosis, and forced vital capacity percent at diagnosis in the analyzed patients.

Figure 5.. The eleven features used in the supervised machine learning model to predict ALS clinical subtype.

(A) Distribution of the Shap values for the eleven features with the most significant impact on the classification model. Each point represents a subject and may have a positive or negative impact depending on its SHAP value. For instance, high values of the rate of BMI decline in red contribute strongly to the positive class, while low values in blue contribute to a lesser extent to the negative class. (B & C) The aggregate of the Shap values is shown for the top eleven features (ranked from most to least important). (D) Model output trajectory for a single subject with the bulbar subtype of ALS. The predicted probability that the patient had the bulbar subtype of ALS was 0.91, predominantly driven by the patient’s bulbar site of symptom onset and only minorly driven by their smoking status and El Escorial category at diagnosis. See https://share.streamlit.io/anant-dadu/machinelearningforals/main for more examples.