Objective:

To characterize the relationship between brain morphometry and the levodopa equivalent daily dose (LEDD) in patients with Parkinson’s Disease (PD). 

Background:

The daily amount of antiparkinsonian medication taken can be estimated via the levodopa equivalent daily dose (LEDD). PD patients typically require increasing LEDD as the disease progresses, however the individual trajectory of medication requirements may be highly variable.

Design/Methods:

We conducted a retrospective study using the Parkinson’s Progression Markers Initiative (ppmi-info.org) and collected the following variables: baseline MRIs, clinical demographics and medication logs. We defined “LEDD delta” as the change in LEDD between the first visit and 5-year follow up visit. Then we performed whole brain morphometry using Freesurfer’s SynthSeg. Using this information, we built an AI model using 60% training, 20% validation and 20% hold out test set design along with a 5-fold cross validation paradigm.   

Results:

76 PD patients were included in our analysis. We defined LEDD delta below the 25th percentile was “stable LEDD” (38 patients) and above the 75th percentile was “progressive LEDD” (38 patients). Stochastic gradient descent-based ML model achieved a mean AUCROC of 70.6% (8.6) across the 5 folds for predicting high versus low LEDD requirements. 11 brain areas were identified as important features.  There was a significant difference in the left acumbens and left caudal middle frontal cortex volumes between stable and progressive LEDD groups. 

Conclusions:

We developed a preliminary model that accurately predicts high versus low LEDD requirements within the first 5 years of disease. Key structures identified include the left acumbens and left caudal middle frontal cortex. Further studies are needed to explore the relationship between these brain regions and levodopa requirements.