We have limited knowledge about Electrocardiogram (ECG) markers of Parkinson’s disease (PD) during the early stage. Our study aimed to create a reliable AI-based ECG model to predict PD risk up to 5 years before the disease develops. We conducted a retrospective case-control study using samples from Loyola University Chicago (LUC) and University of Tennessee-Methodist Le Bonheur Healthcare (MLH). The cases and controls were matched based on specific characteristics such as date, age, sex, and race. Only data available at least 6 months before PD diagnosis was used as input for the model.
The data from LUC covered the period from May 2014, while data from MLH extended to January 2015. We identified PD cases using primary ICD diagnostic codes, namely ICD9 332.0 and ICD10 G20. The prediction focused on prodromal PD, which refers to 6 months to 5 years preceding PD diagnosis. To achieve this, we employed a novel deep neural network using standard 10-second 12-lead ECGs. This model was then compared to various feature engineering-based models. We also conducted subgroup analyses based on gender, race, and age.
The one-dimensional convolutional neural network (1D-CNN) was utilized to predict PD risk or identify prodromal PD from ECGs collected between 6 months to 5 years before clinical diagnosis. The model was developed using MLH data and externally validated using LUC data. In total, we identified 131 cases and 1058 controls at MLH, as well as 29 cases and 165 controls at LUC. The model was trained on 90% of the MLH data, internally validated on the remaining 10%, and externally validated on LUC data.
The best performing model achieved an external validation AUC of 0.67 when predicting prodromal PD within 6 months to 5 years. The accuracy improved when using ECGs to predict prodromal PD within 6 months to 3 years, with an external validation AUC of 0.69. The highest AUC was achieved when predicting PD within 1 year before onset, with a value of 0.74. This predictive model was developed using only raw ECGs as inputs and proved effective in identifying individuals with prodromal PD, especially closer to disease diagnosis. The ECG-based model outperformed multiple models built using ECG feature engineering.
Our research highlights the potential of standard ECGs in cost-effectively identifying individuals with prodromal PD, facilitating early detection, and their inclusion in disease-modifying therapeutic trials. Subgroup analyses indicated that certain groups, such as females and individuals over 60 years of age, may benefit from closer monitoring when symptoms start becoming more evident but are not sufficient for a diagnosis.