EDITORIAL Martin J. McKeown, BEng, MD, FRCP(C) Guerry M. Peavy, PhD Correspondence to Dr. McKeown: martin.mckeown@ubc.ca Neurology ® 2015;84:2392–2393 See page 2413 Biomarkers in Parkinson disease It’s time to combine A biomarker can be defined as “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic process or pharmacological response to a therapeutic interven- tion.” 1 Parkinson disease (PD) is in desperate need for accurate biomarkers: patients with PD typically become symptomatic when more than 50% of dopa- minergic neurons are lost, suggesting a prolonged pre- symptomatic period. Early in the disease course, it can be difficult to differentiate PD from other con- ditions such as essential tremor (ET), multisystem atrophy, and progressive supranuclear palsy. Even with the correct diagnosis, PD has wide variability in disease course, rate of progression, and response to treatment, making an objective marker extremely useful. A wide variety of biomarkers have been proposed for PD, including motor performance tests such as finger tapping, oculomotor measures, electrophysio- logic measures, neuroimaging (MRI, PET, SPECT), transcranial sonography, cardiac metaiodobenzylgua- nidine (MIBG) scintigraphy, metabolomic measures, biochemical tests, and microarrays (see review 2 ). These tests are meant to be more sensitive, reproduc- ible, and quantitative than clinical scales such as the Unified Parkinson’s Disease Rating Scale. Such scales frequently are limited from floor and ceiling effects and other nonlinear relationships between disease severity and scores and, by definition, are unhelpful in the presymptomatic stage. Furthermore, the effects of dopaminergic medication such as L-dopa may con- found interpretation of motor scores and oculomotor measures such as saccades. Transcranial sonography may be a sensitive tool, but hyperechogenicity does not vary during the course of PD, is operator depen- dent, and is seen in a substantial proportion of sub- jects with essential tremor as well. Investment in relatively expensive technology does not ensure highly accurate biomarkers. Metabolomic profiling, whereby many (.1,000) metabolites are probed for characteristics of disease appears promis- ing, but separating core features of the disease with other changes in metabolites due to intersubject variation and effects of treatment is difficult. While functional neuroimaging measures such as PET and SPECT have proved valuable in examining some as- pects of the rate of progression of the disease, their use as sensitive biomarkers has not been definitively es- tablished (e.g., reference 3). Microglial activation as- sessed with specific ligands in PET scanning is a potential candidate as a biomarker, but studies show discrepancies between activation seen in the midbrain and striatum. Impaired cardiac uptake of 123 I-MIBG, suggesting sympathetic involvement in PD, has also been suggested as an early biomarker in PD. Which of the dizzying array of potential PD bio- markers is best? In our view, it is unrealistic to expect that there will be a “magic bullet” that will discrim- inate between PD and other diseases with high sen- sitivity and specificity as well as monotonically track progression throughout the entire disease course. PD is a complex disorder that results in abnormal protein aggregation, 4 disrupted biochemical milieu, 5 abnormal brain oscillations, 6 affected functional con- nectivity, 7 even altered gross brain morphology. 8 Assessment of PD will therefore benefit by probing the brain at different temporal and spatial scales. It is far more likely that a group of biomarkers will need to be judiciously combined in a mathematical model to accurately predict disease status. In the machine learning literature dealing with “Big Data,” such com- bined models are referred to as “ensemble methods.” These analytical methods are principled ways to include “weak” individual biomarkers into a com- bined, robust estimate of disease severity. The indi- vidual biomarkers chosen could depend not only on accuracy of the final combined biomarker but also on cost (e.g., PET imaging is expensive) as well as avail- ability (e.g., some genetic markers may be only avail- able in a small group of highly specialized centers). In fact, one could easily envision a panel of biomarkers whereby each (combined) biomarker prognosticates on different aspects of the disease (e.g., risk of falling, risk of developing dementia). Based on the above discussion, the work by Campbell et al. 9 is an exciting development. By From the Departments of Medicine (Neurology) and Electrical and Computer Engineering (M.J.M.), University of British Columbia, Vancouver, Canada; and Department of Neurosciences (G.M.P.), University of California, San Diego. Go to Neurology.org for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the editorial. 2392 © 2015 American Academy of Neurology ª 2015 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.