A DEEP-LEARNING algorithm using both diagnostic and medication data effectively predicts Parkinson’s disease (PD) in its prodromal phase, with the highest accuracy observed when analysing early prescription patterns.
PD is a neurodegenerative disorder often preceded by a range of prodromal symptoms, some of which are highly specific while others are more common and non-specific. Identifying individuals at risk of developing PD based on these early, non-specific symptoms remains a challenge. Deep-learning techniques offer a promising solution by analysing large datasets to improve predictive accuracy. This study aimed to enhance the performance of deep-learning-based screening for prodromal PD by incorporating medical claims and prescription data.
Using Korean National Health Insurance cohort data, the study sampled 820 patients with PD and 8,200 age- and sex-matched controls. A deep-learning algorithm was trained using combinations of diagnostic codes, medication codes, and different prodromal periods. When analysing data from the 3 years preceding PD diagnosis, prediction using only diagnostic codes achieved an accuracy of 0.937. The addition of prescription data for this period did not significantly improve accuracy (0.931–0.935). However, for an earlier prodromal period (6–3 years before diagnosis), accuracy was lower when using only diagnostic codes (0.890), but significantly improved to 0.922 when medication data was included. These findings suggest that prescription patterns provide valuable predictive information in the earlier stages of prodromal PD.
In clinical practice, early identification of individuals at risk of developing PD is essential for timely intervention and research into disease-modifying treatments. A surveillance system utilising automatically collected medical claims and prescription data could be a cost-effective method for screening those at high risk. This could improve early diagnosis rates and streamline the selection of appropriate candidates for clinical trials, accelerating the development of effective therapies.
Jenna Lorge, EMJ
Reference
Koo Y et al. Predicting Parkinson’s disease using a deep-learning algorithm to analyze prodromal medical and prescription data. J Clin Neurol. 2025 Jan;21(1):21-30.
