Algorithm encompassing clinical and genetic factors predicts PD

By Lucy Piper, Senior medwireNews Reporter

Researchers have devised a model that distinguishes patients with Parkinson’s disease (PD) from those without the condition without relying on motor features.

The algorithm can be administered remotely and “at a fraction of the cost” of dopamine transporter scanning, say researcher Andrew Singleton (National Institutes of Health, Bethesda, Maryland, USA) and colleagues. They believe that the diagnostic model will be useful to researchers in the clinical trial setting, but concede that the low prevalence of PD in the general population restricts its use as a general screening tool.

The researchers developed their algorithm in the Parkinson’s Progression Marker Initiative (PPMI) case-control study, which involved 367 patients with recently diagnosed PD verified by dopamine transporter imaging and 165 neurologically healthy individuals.

Using stepwise logistic regression they removed uninformative variables from a constellation of non-motor features and known PD risk factors.

The resulting model included olfactory discrimination according to the University of Pennsylvania Smell Identification Score (UPSIT), family history of PD, age, gender and a composite genetic risk score based on 30 genetic variants. At its optimum cutoff threshold (receiver operating curve of 0.655), the model correctly identified 83% of patients with PD, at a specificity of 90%.

And it stood up to external validation when tested in a total of 825 PD patients and 261 controls from five independent cohorts with varying recruitment designs and strategies, accurately classifying PD with an area under the curve ranging from 0.89 to 1.00.

Singleton and co-workers note, however, that, although the sensitivity and specificity values seem high, if applied to the general population, for whom the prevalence of PD is low at about 2% for those aged over 60 years, the model would have a low positive predictive value of 15% – falsely identifying six individuals as having PD for every real case identified.

“These data show that the integrative model might be most useful to identify Parkinson’s disease in high-risk populations”, say the researchers.

They tested the model’s ability to identify PD in 55 patients who clinically appear to have typical PD but show no evidence of dopaminergic deficit on imaging scans. The model correctly classified four of five patients who 1 to 2 years later showed evidence of dopaminergic deficit, but also misclassified PD in 13 of 50 patients who had normal scans at follow-up, giving a specificity of 74%.

“Our model was able to discriminate patients without evidence of dopaminergic deficit typical of Parkinson’s disease from those patients with aetiologically typical disease”, the researchers report in The Lancet Neurology.

In a related comment, Samuel Goldman (University of California, San Francisco, USA), discusses the “striking” finding of how well olfactory impairment distinguished between PD cases and controls.

This component accounted for 63.1% of the explained variance in the model and Goldman notes that this component alone “outperformed the classification accuracy of the integrated model in two of the validation populations”. By contrast, the genetic risk score had “only incremental effects on classification accuracy”.

He concludes that this is “an exciting area of research”, but recommends that the current model and future models for predicting PD “be tested using prospective designs that longitudinally follow-up people classified at baseline as high-risk, to see what proportion develop Parkinson’s disease.”

Licensed from medwireNews with permission from Springer Healthcare Ltd. ©Springer Healthcare Ltd. All rights reserved. Neither of these parties endorse or recommend any commercial products, services, or equipment.