Evaluation of longitudinal image-derived AI prognostication as a predictor of overall survival (OS) in a phase 3 advanced non-small cell lung cancer (aNSCLC) trial.
Bibliographic record
Abstract
1552 Background: Confidently anticipating an overall survival (OS) benefit in cancer care and therapeutic research is a defining challenge. AI tools may offer longitudinal measurements that predict OS differences objectively from existing data. Longitudinal imaging-based prognostication (IPRO-Δ), a fully automated deep learning system, was independently trained on real-world imaging data to predict survival from pairs of longitudinal computed tomography (CT) scans. Methods: We retrospectively assessed and compared how IPRO-Δ and percent change in RECIST sum of longest diameters (ΔSLD) predicted OS from 165 pairs of baseline and week 13 CT scans acquired in NExUS (NCT00449033), a phase 3 randomized controlled trial evaluating chemotherapy in combination with either sorafenib or placebo for first-line treatment of subjects with aNSCLC. The two study arms did not show a difference in OS and were combined for this analysis. To examine the association of IPRO-Δ and ΔSLD with OS, we measured the concordance index (c-index) and the standardized hazard ratios (HRs, change in risk for a one-standard-deviation increase in the marker). We also report median OS (mOS) for patients with partial response (PR, n = 60), stable disease (SD, n = 83) and progressive disease (PD, n = 22) at week 13, as defined by RECIST 1.1 guidelines (zero patients had a complete response). To explore the stratification potential of IPRO-Δ, we also define equivalent strata by ordering patients by their IPRO-Δ score and maintaining the same proportions (e.g., the top 60 patients by IPRO-Δ would be IPRO-PR, while the bottom 22 patients would be IPRO-PD), and report the mOS for these strata. Results: For the combined trial arms, median OS was 10.9 months (95% CI: 8.6 – 13.6), 108 (65.4%) were male, and 145 (87.9%) were diagnosed as stage IV. Table 1 reports the c-index, HR and stratified mOS values for both survival markers. Conclusions: At week 13 in NExUS, IPRO-Δ predicted OS differences significantly better than ΔSLD. Future work will explore how IPRO-Δ could serve as the basis for a surrogate endpoint in aNSCLC trials. Summary of association of IPRO-Δ and ΔSLD with OS. IPRO-Δ @ Week 13 (95% CI) ΔSLD @ Week 13 (95% CI) p -value C-Index 0.654 (0.604 – 0.713) 0.543 (0.495 – 0.599) <0.01 HR (1SD) 1.72 (1.38 – 2.15) 1.14 (0.94 – 1.38) <0.01 OS (months, PR / SD / PD or IPRO equivalent) 16.5 / 10.9 / 5.7 12.5 / 12.3 / 4.6 -
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 | 0.004 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".