Fairness audits in clinical Artificial Intelligence convert continuous fairness metrics into binary pass-or-fail verdicts against operational thresholds, where hospital governance boards, payers, and regulators act on the resulting verdicts. Such audits are repeated over time and across hospital sites, thus the same verdict can flip between pass and fail across audits. Existing uncertainty methods such as Bayesian posteriors, bootstrap confidence intervals, and permutation tests address verdict instability only at the continuous-metric level. Converting metric-level uncertainty into a verdict-stability claim remains a manual step that scales poorly across the (model, metric, attribute) cells an audit covers. Existing uncertainty methods also leave open whether bias-mitigation steps, such as reweighing or per-group threshold shifts, yield a stable passing verdict at the cost of model discrimination measured as AUROC or AUPRC.To address this verdict-stability gap, we propose VFR-Audit, a framework built around the Verdict Flip Rate (VFR), a scalar bounded between 0 and 0.5 that measures the probability of verdict reversal under stratified bootstrap resampling. VFR-Audit reports VFR alongside three reliability axes, namely within-cohort resampling stability, audit-size sensitivity, and cross-hospital verdict agreement via Fleiss' kappa.
Fairness audits for LLM-based recommenders have largely focused on observable outputs, implicitly assuming that stable recommendations reflect stable internal processing. We challenge this assumption with FairGap, the first benchmark to jointly evaluate recommendation fairness at two levels: observable output shift (OBS) and hidden representation shift (IBS), measured through controlled counterfactual identity probes across gender, age, and race. Their relationship is summarized via Representation-Output Alignment (ROA), with quadrant diagnostics for identifying user-level hidden-output mismatch. Applied to six open-weight LLM families across three domains, FairGap reveals pervasive hidden-output decoupling: ROA rarely exceeds 0.22, and a non-negligible user population shows stable outputs despite substantial internal shifts, a mode that output-only audits cannot detect by design. Further, activation steering that reduces IBS by up to 8x simultaneously worsens OBS, demonstrating a fundamental tension between internal and output-level fairness that existing frameworks are unequipped to diagnose.
Nitish Nagesh, Elahe Khatibi, Thomas Dean Hughes +3cs.LG
Causal discovery recovers directed structure from observational data and is increasingly used in clinical settings to support mechanism reasoning and fairness audits of predictive models. Path-specific counterfactual fairness asks whether a protected attribute influences an outcome through illegitimate pathways, but these estimands are defined relative to a supplied causal graph and therefore inherit whatever errors the discovery step introduces. Discovery methods are routinely scored on aggregate structural metrics that weight all edges equally, and no established evaluation asks whether the specific pathway an audit depends on survives discovery---or what the audit reports when that pathway is missing. Here we show that full-graph Graph-of-Thoughts reasoning yields acyclic discovered graphs that are structurally competitive with large language model (LLM) baselines, yet that structural fidelity alone does not guarantee fairness-faithful audits. We introduce GoT-CD, in which the reasoning unit is a complete candidate edge set: multiple graphs are generated in parallel, scored by a deterministic validity function, and merged under a hard union constraint that forbids invented edges, with greedy projection enforcing a DAG before commitment. GoT-CD returns a valid DAG on all five reported benchmarks and achieves the best DAG-valid F1 score among LLM methods on Asia, Alzheimer's, and COVID-Respiratory datasets. On an Alzheimer's benchmark with known unfair path, a post-hoc path-specific audit shows that five of eight discovered graphs recover no path from the sensitive attribute to the outcome and therefore report a null overall effect while mediated effects persist, necessitating downstream path-specific fairness analysis along with structural discovery.
Per-subgroup fairness audits of medical image classifiers face a sample-size problem: minority subgroups in held-out test sets have so few samples that the resulting confidence intervals on per-subgroup performance are wider than the bias the audit is meant to detect. We argue that a demographically-conditioned synthetic generator can do both: mitigate bias on the training side and detect bias on the evaluation side. Working on COVID-19 chest CT classification with an end-to-end fine-tuned Stable Diffusion 2.1 generator, we make two findings. For bias mitigation (training), a demographically-balanced synthetic cohort is most useful as a pretraining prior, not as joint augmentation: with the same fixed data, sequential pretraining followed by fine-tuning substantially outperforms joint augmentation, and the resulting classifier surpasses the full-real baseline at $\sim$$100\times$ real-data efficiency. For bias detection (evaluation), across five synthetic minority cohorts and five classifier seeds, the synthetic estimator reproduces the subgroup ranking of a well-powered real oracle (Spearman $ρ= 1.00$ on MCC and Recall) and gives the more reliable per-cell estimate where the small real test set runs out of samples. The synthetic cohort is therefore most useful in exactly the cells that fairness audits care about, as both a fix for and a measure of subgroup bias.
Automated segmentation of cervical-spine MRI is increasingly used in clinical workflows, yet no fairness audit exists for this anatomy. We show that auditing these segmentation tasks is complicated by a common property of modern segmentation datasets: expert-annotated gold labels are expensive, so abundant machine-generated (silver) labels are added to limit annotation cost. This matters because the reference used to judge a model can itself be biased. In this study, we present the first fairness audit of cervical-spine MRI segmentation across sex, age, and race using the CSpineSeg dataset. We observe that the deployed model is demographically fair, but the choice of reference label, however, is not neutral. Because a dataset's silver labels are generated by a model trained on its gold labels, any new model trained on those same gold labels agrees more with the silver labels than with expert truth: scoring identical predictions against silver rather than gold overestimates performance by ~8 Dice points and turns the fairness verdict for age from non-significant to significant - not by the gap inflation Parikh et al. report (which we term false magnitude) but by collapsing within-group variance (which we term false confidence). Reference-label provenance is thus a first-order confounder in segmentation evaluation: performance and fairness should be reported against expert labels, and any fairness claim stated together with the provenance of its reference.
Ha-Hieu Pham, Hai-Dang Nguyen, Dang P. M. Cao +5cs.LG cs.CV
In chest X-ray (CXR) classification, acceptable ranking performance can still leave rare-positive patients below threshold, especially within subgroups. We study this pre-deployment fairness problem as an audit question: after a long-tailed multi-label CXR model is converted from scores into decisions, who is missed? Across VinDr-CXR and MIMIC-CXR/CXR-LT, we use a diagnostic ladder to separate class-level long-tail losses, subgroup-aware weighting, group robustness, and threshold selection. On VinDr-CXR, group-tail weighting followed by tail-aware thresholding reduces tail FNR from 0.665 to 0.269, sex worst-group FNR from 0.705 to 0.157, and age worst-group FNR from 0.822 to 0.133, while macro-mAP increases from 0.611 to 0.635. On MIMIC-CXR/CXR-LT, the same score-to-threshold comparison reduces tail FNR from 0.866 to 0.741 and lowers worst-group FNR across sex, age, race, and insurance; residual missed-positive rates nevertheless remain high. Paired bootstrap contrasts on VinDr support the thresholded FNR reductions, and GroupDRO reference runs indicate that aggregate group robustness alone does not remove rare subgroup misses in this setting. The study supports a narrow audit claim: rare-label fairness in CXR depends jointly on the finding, subgroup, and operating threshold, not on label frequency or ranking metrics alone.
We introduce the Cross-Platform Fairness Evaluation (CPFE) framework -- a five-axis audit protocol covering discriminative performance, calibration, statistical significance, prediction equity, and attribution stability -- and apply it to four transformer models (BERT, RoBERTa, Emotion-DistilRoBERTa, GoEmotions-RoBERTa) trained on a Kaggle mental health corpus (n=35,556) and evaluated on Reddit (n=6,257) and Twitter (n=2,883) test sets with emotion labels mapped to clinical proxies. All three independently evaluated models exhibit consistent and substantial cross-platform AUC degradation (30.3-35.4% on Reddit, 37.9-39.5% on Twitter) relative to within-platform performance (AUC 0.983-0.987), confirmed across five independent training seeds. Calibration failure is concurrent and severe: ECE rises from 0.056-0.060 in-domain to 0.196-0.229 on Reddit and 0.499-0.542 on Twitter. Platform-specific temperature scaling reduces mean ECE by 88.0% without altering discriminative performance (mean |delta AUC|<0.01), confirming separable failure modes. Prediction equity analysis reveals large cross-platform disparities (raw DI < 0.17; prior-shift-adjusted DI: 0.11-0.29 on Reddit), with equalized odds differences of 0.753-0.830 for mental health proxy classes on Reddit and 0.755-0.831 for anxiety on Twitter. Attribution stability analysis shows near-complete vocabulary divergence across platforms (Jaccard J=0 in 14/16 model-class pairs at K=10). These findings support treating cross-platform validation across all five CPFE axes as a standard requirement for mental health NLP systems in heterogeneous environments. In a single-seed fine-tuning experiment, mean AUC improved by 0.216, suggesting target-platform labels provide greater benefit as training signal than as calibration signal.
Second-language (L2) English learners can rarely rehearse speaking with a partner. Speaking is also the most anxiety-laden skill. These gaps drive a fast-growing market for automated speaking practice and scoring. But an automated score is trustworthy only if it is accurate, interpretable, fair, and benchmarked against the right human bar. We build an interpretable feature-plus-LLM hybrid for spontaneous L2 dialogue. We evaluate it without ever fitting to the human labels, against the ICNALE Global Rating Archive: 140 speeches rated by ~80 trained raters on 10 analytic criteria. We score the 130 L2 speeches with usable audio. A deterministic De-Jong speech-timing composite reaches rho=0.764. Blended with a single text-LLM fluency judgment, it reaches Spearman rho=0.818 against the consensus gold. This agrees with the consensus better than 81% of the 80 individual trained raters: above the median rater (rho=0.73) and near the best, and at ~83% of the reliability-corrected maximum (kappa_max=0.99). The blend improves on the composite alone by +0.054 (paired-bootstrap 95% CI [0.017, 0.108], excludes 0); the LLM adds a coarse fluency ranking that the continuous composite refines. We also report a controlled null on pause encoding, bounded to effects below about +/-0.1 rho at this sample size. Holding the LLM and learner words fixed and varying only how pauses are written into the prompt, inline pause locations do not beat aggregate pause statistics (-0.069, CI [-0.15, +0.08]), and a grounded mid-clause criterion gives no reliable gain. The fluency signal comes from the measured speech-timing features, not from how pauses are written for the LLM. We back every claim with two agreeing learner-isolation methods, paired-bootstrap CIs, a monologue negative control, per-feature reproduction of classical measurements, and a per-L1 fairness audit.