Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-0250
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-07-17T22:09:50.064548Z
status: published
lens: g_space
sector: health
headline: Phenotyping antidepressant treatment response with deep learning in electronic health records
dek: ABSTRACT Efficient, accurate phenotyping for antidepressant treatment response in electronic health records (EHRs) could facilitate precision psychiatry applications but remains a challenge. Increasingly, artificial intelligence methods using “deep learning” applied to clinical data have shown promise in complex classification problems. Here, we systematically evaluate the performance of eight deep-learning-based natural language processing models in classifying response to antidepressants in a large real-world…
gain_title: Deep-learning NLP models classified antidepressant treatment response as improved versus no evidence of improvement from routine EHR clinical notes with AUROC up to 0.88.
problem_title: (none)
trace_subject: (none)
gain_reading: Deep-learning NLP models classified antidepressant treatment response as improved versus no evidence of improvement from routine EHR clinical notes with AUROC up to 0.88.
gain_evidence: deep learning methods applied to EHR data can accurately classify antidepressant response in a real-world healthcare setting
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers evaluated eight deep-learning natural language processing models to phenotype antidepressant treatment response from routine clinical notes in the Mass General Brigham system. Using 111,572 patients from 1990-2018 and 4,299 manually reviewed note sets across 2 days to 26 weeks after initiation, models distinguished 'improved' versus 'no evidence of improvement' with strong discrimination.

Accurate automated phenotyping could enable precision psychiatry research and clinical decision support at scale, reducing reliance on manual chart review. As of the July 2026 publication date, results were retrospective and limited to one health system, leaving prospective deployment, generalizability, and impact on patient outcomes still to be demonstrated.
limitation: 
tag: Evidence-backed gain
key_points: Study used EHR data warehouse of the Mass General Brigham healthcare system (n=111,572) spanning 1990-2018 for adults with depression and antidepressant prescription. | Researchers manually reviewed stratified random sample of 4,299 note sets across three windows after initiation: 2 days to 4 weeks, 4-12 weeks, and 12-26 weeks. | Eight deep-learning-based natural language processing models were systematically evaluated, with PPVs ranging from 0.72 - 0.91. | Models incorporating more information-dense and longer text sequences performed better than others.
rundown: The cohort included 111,572 adults with depression and co-occurring antidepressant prescription identified from Mass General Brigham EHR data from 1990-2018. Clinical notes were collected in three post-initiation windows and a stratified random sample of 4,299 note sets was manually labeled for response status.

Eight deep-learning NLP models were tested, all achieving AUROC of at least 0.80 and PPVs between 0.72 and 0.91. The Longformer-large with sliding window was best, and the authors noted that longer, more information-dense text sequences generally improved performance.
sources:
- peer_reviewed | Translational Psychiatry | https://doi.org/10.1038/s41398-026-04266-1 | 2026-07-10
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

cbc2944812dcbc652c8c1c36b4379b1952eda21ea4f5bdd5666c4a5d7346341f

previous

0000000000000000000000000000000000000000000000000000000000000000