9 March 2020 in Nature Medicine
7 January 2025 in Journal of the American Medical Informatics Association
WAWouter A C van Amsterdam
+14
The primary practice of healthcare artificial intelligence (AI) starts with model development, often using state-of-the-art AI, retrospectively evaluated using metrics lifted from the AI literature like AUROC and DICE score. However, good performance on these metrics may not translate to improved clinical outcomes. Instead, we argue for a better development pipeline constructed by working backward from the end goal of positively impacting clinically relevant outcomes using AI, leading to considerations of causality in model development and validation, and subsequently a better development pipeline. Healthcare AI should be “actionable,” and the change in actions induced by AI should improve outcomes. Quantifying the effect of changes in actions on outcomes is causal inference. The development, evaluation, and validation of healthcare AI should therefore account for the causal effect of intervening with the AI on clinically relevant outcomes. Using a causal lens, we make recommendations for key stakeholders at various stages of the healthcare AI pipeline. Our recommendations aim to increase the positive impact of AI on clinical outcomes.
6 July 2024 in npj Digital Medicine
Automatic assessment of impairment and disease severity is a key challenge in data-driven medicine. We propose a framework to address this challenge, which leverages AI models trained exclusively on healthy individuals. The COnfidence-Based chaRacterization of Anomalies (COBRA) score exploits the decrease in confidence of these models when presented with impaired or diseased patients to quantify their deviation from the healthy population. We applied the COBRA score to address a key limitation of current clinical evaluation of upper-body impairment in stroke patients. The gold-standard Fugl-Meyer Assessment (FMA) requires in-person administration by a trained assessor for 30-45 minutes, which restricts monitoring frequency and precludes physicians from adapting rehabilitation protocols to the progress of each patient. The COBRA score, computed automatically in under one minute, is shown to be strongly correlated with the FMA on an independent test cohort for two different data modalities: wearable sensors (ρ = 0.814, 95% CI [0.700,0.888]) and video (ρ = 0.736, 95% C.I [0.584, 0.838]). To demonstrate the generalizability of the approach to other conditions, the COBRA score was also applied to quantify severity of knee osteoarthritis from magnetic-resonance imaging scans, again achieving significant correlation with an independent clinical assessment (ρ = 0.644, 95% C.I [0.585,0.696]).
26 June 2023 in Proceedings of the AAAI Conference on Artificial Intelligence
Methods which utilize the outputs or feature representations of predictive models have emerged as promising approaches for out-of-distribution (OOD) detection of image inputs. However, as demonstrated in previous work, these methods struggle to detect OOD inputs that share nuisance values (e.g. background) with in-distribution inputs. The detection of shared-nuisance OOD (SN-OOD) inputs is particularly relevant in real-world applications, as anomalies and in-distribution inputs tend to be captured in the same settings during deployment. In this work, we provide a possible explanation for these failures and propose nuisance-aware OOD detection to address them. Nuisance-aware OOD detection substitutes a classifier trained via Empirical Risk Minimization (ERM) with one that 1. approximates a distribution where the nuisance-label relationship is broken and 2. yields representations that are independent of the nuisance under this distribution, both marginally and conditioned on the label. We can train a classifier to achieve these objectives using Nuisance-Randomized Distillation (NuRD), an algorithm developed for OOD generalization under spurious correlations. Output- and feature-based nuisance-aware OOD detection perform substantially better than their original counterparts, succeeding even when detection based on domain generalization algorithms fails to improve performance.
13 December 2024 in JMIR Medical Informatics
Background: Prediction models based on machine learning (ML) methods are being increasingly developed and adopted in health care. However, these models may be prone to bias and considered unfair if they demonstrate variable performance in population subgroups. An unfair model is of particular concern in bladder cancer, where disparities have been identified in sex and racial subgroups. Objective: This study aims (1) to develop a ML model to predict survival after radical cystectomy for bladder cancer and evaluate for potential model bias in sex and racial subgroups; and (2) to compare algorithm unfairness mitigation techniques to improve model fairness. Methods: We trained and compared various ML classification algorithms to predict 5-year survival after radical cystectomy using the National Cancer Database. The primary model performance metric was the F1-score. The primary metric for model fairness was the equalized odds ratio (eOR). We compared 3 algorithm unfairness mitigation techniques to improve eOR. Results: We identified 16,481 patients; 23.1% (n=3800) were female, and 91.5% (n=15,080) were “White,” 5% (n=832) were “Black,” 2.3% (n=373) were “Hispanic,” and 1.2% (n=196) were “Asian.” The 5-year mortality rate was 75% (n=12,290). The best naive model was extreme gradient boosting (XGBoost), which had an F1-score of 0.860 and eOR of 0.619. All unfairness mitigation techniques increased the eOR, with correlation remover showing the highest increase and resulting in a final eOR of 0.750. This mitigated model had F1-scores of 0.86, 0.904, and 0.824 in the full, Black male, and Asian female test sets, respectively. Conclusions: The ML model predicting survival after radical cystectomy exhibited bias across sex and racial subgroups. By using algorithm unfairness mitigation techniques, we improved algorithmic fairness as measured by the eOR. Our study highlights the role of not only evaluating for model bias but also actively mitigating such disparities to ensure equitable health care delivery. We also deployed the first web-based fair ML model for predicting survival after radical cystectomy.
1 September 2024 in Machine Learning: Science and Technology
Anomaly, or out-of-distribution, detection is a promising tool for aiding discoveries of new particles or processes in particle physics. In this work, we identify and address two overlooked opportunities to improve anomaly detection (AD) for high-energy physics. First, rather than train a generative model on the single most dominant background process, we build detection algorithms using representation learning from multiple background types, thus taking advantage of more information to improve estimation of what is relevant for detection. Second, we generalize decorrelation to the multi-background setting, thus directly enforcing a more complete definition of robustness for AD. We demonstrate the benefit of the proposed robust multi-background AD algorithms on a high-dimensional dataset of particle decays at the Large Hadron Collider.