DISCOVERY AND VALIDATION OF A NEW CLASSIFICATION OF ANA-RMDS THAT BETTER PREDICT LONG-TERM OUTCOMES COMPARED TO LEGACY DIAGNOSES
Bibliographic record
Abstract
O015 / #174 Topic: AS20 - Precision Medicine ABSTRACT CONCURRENT SESSION 02: SLE METRICS – IMPROVING OUTCOMES & MEASURES 22-05-2025 1:40 PM - 2:40 PM Background/Purpose ANA-associated RMDs (ANA-RMDs) include SLE, Sjogren’s, Scleroderma, Myositis, and mixed/undifferentiated CTD. Despite overlapping clinical and immunophenotypic features, there is significant disparity in access to targeted therapies across ANA-RMDs. A robust data-driven reclassification using clinical and biomarker data with clinical impact could define more homogeneous cohorts for therapies and clinical trials. Methods We trained a variational autoencoder with the European PRECISESADS cohort of 876 ANA-RMD patients using R, keras, and tensorflow. 25 covariates were prioritized by ANA-RMD specialists and patient focus groups. Data was compressed to an 8-neuron latent space and analyzed with multiple clustering techniques. For validation, K-means centroids from PRECISESADS were applied to the DEFINITION dataset (219 patients) (Figure 1). Cluster durability was assessed using entropy, elbow plots, and cluster stability index. Gene expression data was analyzed with heatmaps and summary statistics. Clinical impact in DEFINITION was analyzed cross-sectionally and longitudinally using descriptive statistics, PROs (eg, SF36), physician assessments (eg, BILAG-2004, PGA), and gene expression scores. 5-year follow-up outcomes included hospitalization rates. Kaplan-Meier and Sankey plots were generated with survival and flipPlots R packages. Figure 1. Results Deep learning revealed 5 distinct ANA-RMD classes. Each class encompassed patients from various legacy diagnoses, with no single legacy diagnosis mapping to a new class. These classes were: (i) Sicca, mostly patients with a legacy diagnosis of pSS, SLE, or UCTD with low disease activity but high IFN-I expression (Figure 2); (ii) Quiescent, characterized by low gene expression and physician-assessed disease activity but high patient-reported pain scores; (iii) Active MSK disease, with high MSK disease activity and high inflammatory gene expression; (iv) Polyinflammatory, with high levels of therapeutic change, PRO impact, and high myeloid/interferon/inflammatory gene expression, containing substantial numbers of previously undifferentiated patients; (v) Myeloinflammatory, with high healthcare utilization, physician-assessed disease activity, and emergency department attendance (Figure 3). Five-year healthcare data revealed significant differences in hospital admission rates (p<0.01) and emergency department attendance (p<0.01) for the new classes but not for legacy diagnoses. Figure 2. Figure 3. Conclusions Using advanced deep learning, we developed and validated a new classification for ANA-RMDs. Our findings showed that (i) more of the ANA-RMD spectrum could be classified than with legacy diagnoses; (ii) immunophenotypic and clinical features within these classes were more homogeneous than with legacy diagnoses, suggesting suitability for the same therapies and outcomes; (iii) these classes better predicted long-term outcomes and healthcare utilization. Clinical trials in these populations may yield larger effect sizes and provide evidence applicable to more patients, thereby reducing healthcare inequality.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".