Objective: We developed a machine learning (ML) framework that facilitates prediction of individual-level Huntington’s disease (HD) stage. The proposed meta-model approach classifies participants according to binary HD disease stage (pre-manifest HD (PreHD); manifest HD) & also fine-grained disease stage. This is the first work to use such an approach for patient stratification in HD.

Background: Early classification of individual disease stage in neurodegenerative diseases like HD is crucial for patient stratification. ML methods present an opportunity to make patient-specific predictions of disease stage using clinical data, & thus have the potential to advance clinical decision support. However, as yet there has been no attempt to establish a computerised prediction system in HD.

Method: We trained & evaluated 9 different ML models (base models) using baseline cross-sectional data from 184 HD gene-positive participants from TRACK-HD dataset, which provides clinical, imaging & genetic data. Performance was assessed using k-fold cross validation. The predictions of the base models on out-of-sample data were then used to train a separate meta-model, which combines the predictions of base models with the intent of reducing the variance & generalisation error. We hypothesised that a meta-model approach would perform more accurately than the best base model.

Results: The base models individually were able to distinguish PreHD from HD participants with mean accuracy varying within 86%^±8.2 – 96%^±4.6. They classified each participant according to their fine-grained disease stage with mean accuracies varying between 61% ^±12.5 – 82% ^±7.4. In comparison, the meta-model achieved an accuracy of 96.7%^±4.6 in classifying PreHD & HD participants. Our results are promising compared with the accuracy of existing models [1-4] that vary between 75% – 94% on the binary classification task. Further, the meta-model was able to classify participants according to their fine-grained disease stage with an accuracy of 86% ^±7.5, performing significantly better (p < 0.001) than the next most competitive base model.

Conclusion: The meta-model achieved the best predictive accuracy for both the binary and fine-grained classification task. These models are thus a powerful tool for early classification of fine-grained HD disease stage, with potential applications for clinical trial stratification, & can be extrapolated to other neurodegenerative diseases.

References: [1.] Miranda, Â., Lavrador, R., Júlio, F. et al. Classification of Huntington’s disease stage with support vector machines: A study on oculomotor performance. Behav Res 48, 1667–1677 (2016). https://doi.org/10.3758/s13428-015-0683-z [2.] Perez, M., Jin, W., Le, D., Carlozzi, N., Dayalu, P., Roberts, A., & Provost, E. M. (2018). Classification of huntington disease using acoustic and lexical features. In Interspeech (Vol. 2018, p. 1898). NIH Public Access. [3.] Acosta-Escalante, F. D., Beltrán-Naturi, E., Boll, M. C., Hernández-Nolasco, J. A., & García, P. P. (2018). Meta-classifiers in Huntington’s disease patients classification, using iPhone’s movement sensors placed at the ankles. IEEE Access, 6, 30942-30957. [4.] Romana, A., Bandon, J., Carlozzi, N., Roberts, A., & Provost, E. M. Classification of Manifest Huntington Disease using Vowel Distortion Measures. (2020). arXiv preprint arXiv:2010.08503.

« Back to MDS Virtual Congress 2021

MDS Abstracts - https://www.mdsabstracts.org/abstract/fine-grained-prediction-of-huntingtons-disease-stage-using-a-meta-model-approach/