Artificial intelligence may aid in predicting individual outcomes for patients with spinal muscular atrophy, according to a recent study using predictive models with XGBoost algorithm reports that was published in PLOS ONE.
Patients with SMA experience worsening, improving, or no change in symptoms throughout their clinical course. Each individual’s clinical future depends on age as well as several other variables.
The authors of this study created a predictive model and report. They studied the disease’s natural history in 1 center, assessing how clinical variables affected the Hammersmith Functional Motor Scale Expanded (HFMSE) score predictions at 6 and 12 months. Inclusion criteria in this study were an age of less than 2.5 years, SMA type 2 diagnosis with onset between 6 and 18 months of age, and the inability to stand or walk.
Thirty patients and 324 visits passed the filtering criteria. The range of patients’ visits was 3 to 23, with a mean of nearly 11 visits per patient. Regarding the machine model, 22 patients with 239 visits and 8 patients with 85 visits were assigned to the training and testing sets, respectively.
Read more about SMA prognosis
SMA type 2 clinical progression is not a straightforward phenomenon, and although functional measures like HFMSE and others overly emphasize age to capture slopes in the progression, these still sustain variability. That is why this study aimed to train a model capable of predicting disease progression based on individual characteristics.
“Following the suggestion that a composite and individualized model may improve the prognostic accuracy of disease progression, we applied a machine learning approach using an XGBoost algorithm for regression,” the authors wrote. “The advantage and peculiarity of this method compared to multivariate analysis is that it provides an estimate of the possible individual trajectory based on the baseline features, each of them assessed to establish their prognostic value.”
Furthermore, the authors explained that despite age being a crucial risk factor for predictiveness, other variables are also important and were identified in the study, including the HFMSE value at the first visit, age of symptom onset, body mass index, Cobb’s angle, and noninvasive ventilation.
In conclusion, the authors found this model to provide a more individual assessment of the clinical progression of these patients. They suggest applying this study to a larger group of patients to corroborate the findings or even to identify new progression patterns.
Coratti G, Lenkowicz J, Patarnello S, et al. Predictive models in SMA II natural history trajectories using machine learning: a proof of concept study. PLoS One. 2022;17(5):e0267930. doi:10.1371/journal.pone.0267930