Search

Objective: We develop a deep learning framework based on the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model using unstructured clinical notes from electronic health records (EHRs) to predict the risk of disease progression from Mild Cognitive Impairment (MCI) to Alzheimer's Disease (AD). Materials and Methods: We identified 3657 patients diagnosed with MCI together with their progress notes from Northwestern Medicine Enterprise Data Warehouse (NMEDW) between 2000-2020. The progress notes no later than the first MCI diagnosis were used for the prediction. We first preprocessed the notes by deidentification, cleaning and splitting, and then pretrained a BERT model for AD (AD-BERT) based on the publicly available Bio+Clinical BERT on the preprocessed notes. The embeddings of all the sections of a patient's notes processed by AD-BERT were combined by MaxPooling to compute the probability of MCI-to-AD progression. For replication, we conducted a similar set of experiments on 2563 MCI patients identified at Weill Cornell Medicine (WCM) during the same timeframe. Results: Compared with the 7 baseline models, the AD-BERT model achieved the best performance on both datasets, with Area Under receiver operating characteristic Curve (AUC) of 0.8170 and F1 score of 0.4178 on NMEDW dataset and AUC of 0.8830 and F1 score of 0.6836 on WCM dataset. Conclusion: We developed a deep learning framework using BERT models which provide an effective solution for prediction of MCI-to-AD progression using clinical note analysis.

Additional file 4: Figure S4. Loss of APP immunoreactivity in GABABR1-positive interneurons in Aplp2−/−App flox/flox;Gad2-Cre mice. Representative confocal images of mouse hippocampal sections of Aplp2−/−App flox/flox;Gad2-Cre.

Additional file 1: Figure S1. APP mRNA expression in excitatory and inhibitory neurons of the hippocampus. Fluorescent in situ hybridization for App (red), the inhibitory neuron marker Vgat (green) and the excitatory neuron marker Vglut1 (blue) mRNAs in wild type and APP KO coronal sections. Insets show CA1 region. Scalebar is 100 μm.

Additional file 2: Figure S2. Workflow for quantification of Aβ plaque load across hippocampal laminae. Outline of methods developed to quantify Aβ load across hippocampal laminae by segmenting hippocampal laminae, refining hippocampal subfields, creating masks to define Aβ plaque area, and combing the regions of interest (ROIs) for each subfield and laminae with masks for Aβ plaque area.

Additional file 5: Figure S5. Plaque load of 6mo AppNL-G-F/NL-G-F; Gad2-Cre; Bace1flox/flox mice compared to controls. Quantification of the Aβ plaque load in hippomcapus as determined by IHC. Graphs show means ± SEM. (n = 3 mice).

Additional file 3: Figure S3. Laminar distribution of Aβ plaques in the hippocampus of APP knock-in mice. Graphs detailing the data summarized in Fig. 2c-f. Quantification of the Aβ plaque load in hippocampal subfields and laminae of CA1, CA2/3 and dentate gyrus (DG) of sections from 2, 3, 4, and 6 month old APPNL-G-F/NL-G-F mice. Graphs show means ± SEM. (n = 5–7 mice; one-way ANOVA). *P