[home] [Personal Program] [Help]

---

Session: NeuroMuscular
Session starts: Thursday 26 January, 14:30
Presentation starts: 14:45
Room: Room 530

---

Ahnjii ZhuParris (CHDR)
Eva Thijssen (CHDR)
Willem Elzinga (CHDR)
Soma Makai-Boloni (CHDR)
Wessel Kraaij (LIACS)
Geert Jan Groeneveld (CHDR)
Robert-Jan Doll (CHDR)

Abstract:
Background Validation of objective biomarkers to monitor treatment effects among Parkinson’s Disease (PD) patients would benefit anti-parkinsonian drug development. We investigated the performance of machine learning-derived composite biomarkers using finger tapping tasks features to classify treatment effects and to estimate motor symptom severity among PD patients. Methods In a placebo-controlled, cross-over design study, data were collected from 20 PD patients. The Alternative Index Finger Tapping (IFT), Alternate Index and Middle Finger Tapping (IMFT), and Thumb-Index Finger Tapping (TIFT) tasks, and the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) III assessments were performed during the placebo and levodopa/carbidopa treatments. We compared the performance of various composite biomarkers to classify the treatment classes and to estimate the MDS-UPDRS III total scores. The composite biomarkers represent each of the tapping tasks individually, the tapping tasks collectively, and the MDS-UPDRS III item scores. For each composite biomarker, we considered the baseline-uncorrected and corrected values. We compared the performance of three classification models to discriminate between the placebo and active treatments. We used Linear Regression (with Elastic net regularization) to estimate the MDS-UPDRS scores. We used repeated nested cross-validation to build and validate each of the models. For both the classification and regression models, we trained the models on data collected during peak drug concentration, and used the trained models to predict the treatment class probability and the MDS-UPDRS III scores from baseline to the last timepoint. Results The IFT composite biomarker (baseline-corrected) achieved the highest classification performance with 83.50% accuracy and 93.95% precision. The IFT composite biomarker (baseline-corrected) also achieved the optimal regression performance with a Mean Absolute Error of 7.87 points, and a Pearson’s correlation of 0.69. Together, these models demonstrate that the baseline-uncorrected features outperform that of the baseline-corrected features. Conclusion We demonstrated that the IFT composite biomarker outperformed both the 3 tapping task and the MDS-UPDRS III composite biomarkers for the classification of treatment effects. However, the tapping composite biomarkers were not suitable for estimating the MDS-UPDRS III scores. To conclude, the tapping composite biomarkers are suited to complement, but not replace, the clinical assessments of anti-Parkinsonian drug effects.