Assessing Algorithmic Fairness With a Multimodal Artificial Intelligence Model in Men of African and Non-African Origin on NRG Oncology Prostate Cancer Phase III Trials
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
PURPOSE Artificial intelligence (AI) tools could improve clinical decision making or exacerbate inequities because of bias. African American (AA) men reportedly have a worse prognosis for prostate cancer (PCa) and are underrepresented in the development genomic biomarkers. We assess the generalizability of tools developed using a multimodal AI (MMAI) deep learning system using digital histopathology and clinical data from NRG/Radiation Therapy Oncology Group PCa trials across racial subgroups. METHODS In total, 5,708 patients from five randomized phase III trials were included. Two MMAI algorithms were evaluated: (1) the distant metastasis (DM) MMAI model optimized to predict risk of DM, and (2) the PCa-specific mortality (PCSM) MMAI model optimized to focus on prediction death in the presence of DM (DDM). The prognostic performance of the MMAI algorithms was evaluated in AA and non-AA subgroups using time to DM (primary end point) and time to DDM (secondary end point). Exploratory end points included time to biochemical failure and overall survival with Fine-Gray or Cox proportional hazards models. Cumulative incidence estimates were computed for time-to-event end points and compared using Gray's test. RESULTS There were 948 (16.6%) AA patients, 4,731 non-AA patients (82.9%), and 29 (0.5%) patients with unknown or missing race status. The DM-MMAI algorithm showed a strong prognostic signal for DM in the AA (subdistribution hazard ratio [sHR], 1.2 [95% CI, 1.0 to 1.3]; P = .007) and non-AA subgroups (sHR, 1.4 [95% CI, 1.3 to 1.5]; P < .001). Similarly, the PCSM-MMAI score showed a strong prognostic signal for DDM in both AA (sHR, 1.3 [95% CI, 1.1 to 1.5]; P = .001) and non-AA subgroups (sHR, 1.5 [95% CI, 1.4 to 1.6]; P < .001), with similar distributions of risk. CONCLUSION Using cooperative group data sets with a racially diverse population, the MMAI algorithm performed well across racial subgroups without evidence of algorithmic bias.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it