This tool is not a clinical diagnostic device. It is a tool for research use, to explore the relative patterns of cognitive features in different atypical parkinsonian syndromes.
The algorithm z-scores each predictor, then subtracts the mean of the features,
before running a simple logistic classifier.
This means that the classifier focuses on the relative pattern
of features. This minimises the possibility
that the classifier is picking up on just disease severity.
More info...
These classifiers are trained on N=935 PD, 100 PSP, 50 CBS, 53 MSA.
Features were z-scored across people, and then each person's mean across all features
was subtracted from each of their feature values,
so that the classifier only detects the relative profile across features,
ignoring overall severity of symptoms.
The classifier is not corrected for motor symptoms, which correlate
with some of the features more than others.
Atypical PD patients had ACE, PD patients had MoCA from which the ACE was imputed. PD, PSP and CBS patients had UPDRS-I nonmotor features, and patients with MSA had UPDRS imputed from CBI.
Missing features are treated as having zero evidence contribution. Note that this is not the optimal classifier for missing data, since the features are typically correlated.