AI Science Tracker
EN

Generative embedding distinguishes recorded two-year depression trajectories

University of Zurich; ETH Zurich; Radboud University; King’s College London; University of Melbourne; Amsterdam UMC; Leiden University; University of Groningen · 17 Feb 2020 · medicine

Discovery

Generative embedding distinguishes recorded two-year depression trajectories

Source check: Source checked. Not independent scientific confirmation.

Finding: In a retrospective reanalysis of 85 NESDA patients, a support-vector classifier using modeled brain connectivity distinguished chronic depression from rapid remission with 79% balanced accuracy in nested cross-validation. The February 2020 journal analysis follows a September 2019 preprint. Small groups, reused data and uncontrolled treatment limit clinical interpretation.

Technology: Linear support vector machine with dynamic-causal-model generative embedding

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
A learned SVM classified later clinical-trajectory labels using posterior connectivity parameters from a conventional generative model; the AI contribution is the discriminative classifier, not an assumption that causal-model terminology proves causation.
Test and result
Baseline face-processing fMRI and available two-year follow-up identified 39 rapid-remission, 31 gradual-improvement and 15 chronic-course patients. SVM hyperparameters were optimized using training-only inner five-fold cross-validation inside leave-one-person-out evaluation. Chronic versus rapid-remission classification reached 79% balanced accuracy and ROC AUC .87; gradual versus rapid remission reached 61% but did not survive the stated multiple-comparison correction, and chronic versus gradual improvement was at chance.
Limitation / still unknown
This is retrospective prognostic modeling in a selected 85-person cohort, not prospective evaluation of a frozen model. The chronic group comprised only 15 patients; age differed and source-reported age adjustment reduced balanced accuracy to 74%. Treatments and intervening life events were not standardized. An earlier Schmaal analysis used the same source cohort and clinical-trajectory grouping, so this reanalysis is not independent confirmation or first discovery of those trajectories. Source footnotes warn that McNemar comparisons between cross-validated classifiers may be optimistic. The original preprint API dates version1 to September 18, 2019; only its metadata/abstract were available in this review, so journal-specific metrics are not backdated. No individualized-treatment benefit or causal neural mechanism is established.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC7082217/

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: Original primary computational training, clinical/sample definition, represented Results and bounded Discussion/limitations actually read; earlier primary preprint API metadata/abstract read, not its full tables..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 1/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse medicine discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.