The classification system may help in selecting treatment for early-stage patients and identifying participants for clinical trials.
A new classification algorithm involving metabolic brain imaging and pattern analysis can differentiate between various early-stage parkinsonian disorders very accurately, according to the results of a study published in the February Lancet Neurology.
Chris C. Tang, MD, PhD, of the Feinstein Institute for Medical Research, Manhasset, New York, and colleagues sought to improve the early diagnosis of different parkinsonian disorders—and, thus, the prognoses, treatments, and drug trials for such disorders.
The researchers developed an algorithm, involving voxel-based spatial covariance mapping and fluorine-18-labeled-fluorodeoxyglucose-PET images, to classify such disorders as idiopathic Parkinson’s disease, multiple system atrophy, or progressive supranuclear palsy.
Between January 1998 and December 2006, the researchers used their algorithm to classify the conditions of patients with parkinsonian features but uncertain clinical diagnoses. Then they compared these classifications to the final clinical diagnoses made by blinded movement disorders specialists after patient assessments that lasted a mean of 2.6 years. The researchers also tested the classifications’ reproducibility by repeating them in one subgroup of patients, and they compared the classifications with both the final clinical diagnoses and with postmortem diagnoses in another subgroup of patients.
A total of 167 patients had imaging classifications; at least six months of follow-up; and final clinical diagnoses of idiopathic Parkinson’s disease, multiple system atrophy, or progressive supranuclear palsy. Among these patients, the final clinical diagnoses indicated that the imaging classifications of idiopathic Parkinson’s disease had 84% sensitivity, 97% specificity, 98% positive predictive value (PPV), and 82% negative predictive value (NPV). The classifications of multiple system atrophy had 85% sensitivity, 96% specificity, 97% PPV, and 83% NPV. In addition, the classifications of progressive supranuclear palsy had 88% sensitivity, 94% specificity, 91% PPV, and 92% NPV.
Also, among patients who had experienced parkinsonian symptoms for two years or less and who had clinical follow-up for more than two years, the classifications had 100% sensitivity, specificity, PPV, and NPV for both multiple system atrophy and progressive supranuclear palsy. Of the 22 patients who were given repeat imaging classifications, 21 were given the same classification both times. Nine patients had postmortem diagnoses, and six of these diagnoses were the same as both the imaging classification and the final clinical diagnosis.
The researchers concluded that their classification algorithm “produced accurate differential diagnosis.” They noted that, at present, initial diagnoses of idiopathic Parkinson’s disease have only about 75% accuracy, and initial clinical diagnoses of early multiple system atrophy and progressive supranuclear palsy have less than 60% sensitivity. Their algorithm can “help in selecting treatment for early-stage patients and identifying participants for clinical trials,” they said.
In an accompanying editorial, Angelo Antonini, MD, of the IRCCS San Camillo, Venice, Italy, and Parkinson Institute in Milan, predicted that the latter function is where the algorithm will find its “natural application.” He estimated that 15% of patients enrolled in a recent trial on rasagiline and Parkinson’s disease probably did not have the disease.
Dr. Tang and colleagues noted that their algorithm cannot yet be used to differentiate forms of parkinsonism other than the three forms dealt with in their study. But it will be able to do so, the authors said, once specific patterns for other forms are identified, validated, and included.
—Jack Baney