Back to Research

Poster · 2021

Maximising UPDRS information while minimising assessment time

G. Morinan, J. Tang and J. O’Keeffe

Objective

The gold standard for measuring disease severity in PD (Parkinson’s Disease) is the MDS-UPDRS (Movement Disorder Society – Unified PD Rating Scale).1 However, the assessment is complex and time consuming, limiting its deployability in clinical practice and research. We aim to simplify the assessment methodology, while maximising the clinical information that is captured. This promises to facilitate applications such as remote, asynchronous home assessment, by reducing the burden on patients while still enabling high quality clinical management.

Background

The MDS-UPDRS consists of 50 questions and motor examinations, typically taking around 30 minutes to complete. It is known that the UPDRS contains redundancy and could be shortened without significant loss of information.2 Previous work proposed an 8-item simplification based upon clinical expertise.3

Methods

Through an exhaustive computational search, we identified an 8-item subset that minimises information loss when compared to the 50 items, while being constrained to not include items that may be difficult for a home assessment (such as gait, which requires a long corridor). Information loss was measured by the correlation between the rating estimated by a linear regression model that utilised the 8-item scores, and the full assessment rating. This used a dataset of 7,566 assessments, collected through the Parkinson’s Progression Markers Initiative.

Figure 1: two stacked histograms of total MDS-UPDRS rating. The PPMI training data of 7,566 assessments peaks around a rating of 35 to 45 and tails off beyond 100; the NMS validation data of 377 assessments is flatter and broader, peaking around 40 with a secondary cluster near 60 and a tail extending to about 130.
Figure 1. Distribution of MDS-UPDRS score (total of all ratings for parts I, II, III and IV) within the training and validation datasets.
Figure 2: two histograms across all 215,553,195 candidate 8-item subsets. Mean squared error is right-skewed with a mode around 128 to 129 and a tail past 400; the Pearson correlation coefficient is left-skewed with a mode around 0.85 to 0.86 and very few subsets below 0.6.
Figure 2. Histograms showing the distributions of each metric for each of the 8-item subsets (n = 215,553,195), computed on the training dataset. (MSE modal bin: 128–129, Pearson modal bin: 0.85–0.86.)

Results

The 8-item subset identified contained three motor examination items (3.2 Facial Movement, 3.5 Hand Movements, 3.7 Toe Tapping) and five questions (1.13 Fatigue, 2.5 Dressing, 2.10 Tremor, 2.12 Walking and Balance, 4.3 Time Spent in the Off State). Correlation between the 8-item rating and the full 50-item UPDRS rating was highly significant (Pearson’s r = 0.920, p-value < 0.001). This was validated on an unseen dataset of 377 assessments, collected by the MDS. For this dataset the correlation between the 8-item rating and the full 50-item UPDRS rating was also highly significant (Pearson’s r = 0.90, p-value < 0.001).

Figure 3: boxplots of total MDS-UPDRS rating against the best 8-item subset rating from 0 to 32, for training and validation data side by side. The medians rise steadily and near-linearly from about 5 at a subset rating of 0 to about 150 at a subset rating of 32, with the two datasets overlapping closely throughout.
Figure 3. Boxplots showing the distribution of the sum rating of the total 50-item MDS-UPDRS rating for each value of the best 8-item subset rating.
Figure 4: scatter plots of actual against predicted UPDRS total score for the training and validation datasets, with points clustered tightly along the marked x equals y line across the full range from 0 to about 175.
Figure 4. Correspondence between ground truth (actual 50-item MDS-UPDRS total rating) and predicted value (using a linear regression model). The X = Y line is marked in grey.
Figure 5: residual histograms for the training and validation datasets, both approximately symmetric and centred on zero, with the bulk of residuals falling between minus 20 and plus 20 points of total MDS-UPDRS rating.
Figure 5. Residuals of the estimator trained to predict total MDS-UPDRS rating using the ratings of the 8-item subset.

Conclusions

We demonstrate an 8-item subset identified using computational statistics retains the vast majority of information contained within the full 50-item UPDRS. Further work will involve implementing digital tools that enable remote simplified UPDRS assessment using this 8-item subset. We estimate that for the vast majority of PD patients this assessment can be completed in 5–10 minutes.

References

  1. Goetz, C. G. et al. Movement Disorder Society-sponsored revision of the Unified Parkinson's Disease Rating Scale (MDS-UPDRS): scale presentation and clinimetric testing results. Movement Disorders 23, 2129–2170 (2008).↑
  2. Van Hilten, J. J. et al. Rating impairment and disability in Parkinson's disease: evaluation of the Unified Parkinson's Disease Rating Scale. Movement Disorders 9, 84–88 (1994).↑
  3. Hauser, R. A. et al. The UPDRS-8: a brief clinical assessment scale for Parkinson's disease. International Journal of Neuroscience 122, 333–337 (2012).↑

Poster Video

About this poster

This research was originally presented as a poster by Machine Medicine Technologies. This HTML version has been reformatted for accessibility and web presentation, while preserving the content of the original poster.