Abstract Details

ET is among the most prevalent neurological diseases. Diagnosis is based entirely on neurological evaluation. Historically, there were few postmortem brain studies, hindering attempts to develop pathologically based criteria to distinguish ET from control brains. However, an intensive effort to bank ET brains over recent years has resulted in postmortem studies involving >200 brains, which have identified numerous degenerative changes in the ET cerebellar cortex. Although ET and controls have been compared with respect to individual metrics of pathology, there has been no overarching analysis to derive a combination of metrics to distinguish ET from controls.

We asked whether there is a constellation of pathological findings that separates essential tremor (ET) from controls, and how well that constellation performs.

Analyses included 100 ET brains from the Essential Tremor Centralized Brain Repository and 50 control brains. A standard tissue block from the cerebellar cortex was used to quantify 11 metrics of pathological change that reflected changes in the Purkinje cell and related neuronal populations. Three supervised classification algorithms were investigated, including logistic regression with ridge penalty, random forest and gradient boosted decision tree. Data were divided into training and validation samples.

All three algorithms performed similarly to correctly predict a diagnosis of ET. Using logistic regression with a ridge penalty algorithm, the simplest method, sensitivity ranged from 87.5% to 96.64%, with six of eight values >95%, and the specificity ranged from 92.87% to 98.11%, with five of eight values >95%. We also provide a web-based application that uses these metric values, and based on specified cut-offs, determines the likely diagnosis.

These analyses illustrate the ability to correctly predict a diagnosis of ET and set the stage for use of pathologically based criteria to distinguish clinically diagnosed ET cases from controls at the time of postmortem.