Dynamic Bayesian networks for stratification of disease progression in amyotrophic lateral sclerosis

Abstract

Background and purpose Progression rate is quite variable in amyotrophic lateral sclerosis (ALS); thus, tools for profiling disease progression are essential for timely interventions. The objective was to apply dynamic Bayesian networks (DBNs) to establish the influence of clinical and demographic variables on disease progression rate. Methods In all, 664 ALS patients from our database were included stratified into slow (SP), average (AP) and fast (FP) progressors, according to the Amyotrophic Lateral Sclerosis Functional Rating Scale Revised (ALSFRS‐R) rate of decay. The sdtDBN framework was used, a machine learning model which learnt optimal DBNs with both static (gender, age at onset, onset region, body mass index, disease duration at entry, familial history, revised El Escorial criteria and C9orf72) and dynamic (ALSFRS‐R scores and sub‐scores, forced vital capacity, maximum inspiratory pressure, maximum expiratory pressure and phrenic amplitude) variables. Results Disease duration and body mass index at diagnosis are the foremost influences amongst static variables. Disease duration is the variable that better discriminates the three groups. Maximum expiratory pressure is the respiratory test with prevalent influence on all groups. ALSFRS score has a higher influence on FP, but lower on AP and SP. The bulbar sub‐score has considerable influence on FP but limited on SP. Limb function has a more decisive influence on AP and SP. The respiratory sub‐score has little influence in all groups. ALSFRS‐R questions 1 (speech) and 9 (climbing stairs) are the most influential in FP and SP, respectively. Conclusions The sdtDBN analysis identified five variables, easily obtained during clinical evaluation, which are the most influential for each progression group. This insightful information may help to improve prognosis and care.

Digital Alternative Communication for Individuals with Amyotrophic Lateral Sclerosis: What We Have

“…ALS is considered rare, and, despite efforts to seek digital health solutions, there are still significant challenges to be tackled: these include the need for more data, studies, and evidence on disease incidence and prevalence, which are essential but scarce pieces of information in the context of global health [ 12 , 87 , 88 , 89 , 90 , 91 , 92 , 93 , 94 , 95 , 96 , 97 , 98 , 99 , 100 , 101 , 102 , 103 , 104 , 105 , 106 , 107 ]. There are few records or epidemiological studies in Brazil, and only two studies at national level have been mentioned in the scientific literature.…”

ALSFRS-R-SE: an adapted, annotated, and self-explanatory version of the revised amyotrophic lateral sclerosis functional rating scale

“…While digital self-assessment may increase data density by allowing for more frequent capture [44][45][46] and while the ALSFRS-R-SE unifies the structure of individual items, there will still be functional changes not covered by the score. Depending on progression rate and phenotype, the sub-scores of the ALSFRS-R will have a different impact on the total score and prognostic models, prompting the suggestion to focus statistical analysis on these sub-scores rather than the total score [26,47]. Another factor constraining the validity of a self-reported score can be cognitive deficits or affective changes.…”

Triclustering-based classification of longitudinal data for prognostic prediction: targeting relevant clinical endpoints in amyotrophic lateral sclerosis

“…Recent years have witnessed an increasing awareness of the potentialities of machine learning amongst ALS researchers, leading to several applications to ALS cohort data [12][13][14][15][16][17][18][19][20][21] . The great potential of learning stratification models has also shown opportunities for future clinical trials, besides promoting more accurate and trustable predictions by learning group-specific prognostic models 13,[22][23][24] .…”