Abbreviated MDS-UPDRS for Remote Monitoring in PD Identified Using Exhaustive Computational Search

Figure 2

A linear regression model was used to estimate the total 50-item MDS-UPDRS rating (see Section 2.4), using the 8-item rating of the “remote” subset (see Table 1). (Upper left) The correspondence between ground truth and estimations for the training dataset, and (upper center) the validation dataset. Note: the X = Y line is marked in grey. For both datasets, the correlations between ground truth and estimations were highly significant (-value <0.0001), and (upper right) the 95% credible intervals of these correlations overlapped. This suggests the ability of this 8-item subset to estimate the total 50-item MDS-UPDRS rating generalises across different datasets. (Mid left) The residuals of the estimator for the training dataset, and (mid center) the validation dataset. The mean residuals for both datasets were close to zero, and (mid right) the 95% credible intervals of the mean residual crossed zero for both datasets. This suggests the linear regression model is an unbiased estimator of the total 50-item MDS-UPDRS rating. (Lower left) The Bland–Altman plot of agreement between the actual and estimated total score for training (lower center) and validation, both of which show that residuals are approximately equally distributed either side of zero across the range of possible values of total score, indicating a lack of proportional bias.