Analyzing the role of model uncertainty for electronic health records

By Michael W. Dusenberry, Dustin Tran, Edward Choi, Jonas Kemp, Jeremy Nixon, Ghassen Jerfel, Katherine A. Heller, Andrew M. Dai. 2019

123 citations recorded by the source corpus.

In medicine, both ethical and monetary costs of incorrect predictions can be significant, and the complexity of the problems often necessitates increasingly complex models. Recent work has shown that changing just the random seed is enough for otherwise well-tuned deep neural networks to vary in their individual predicted probabilities. In light of this, we investigate the role of model uncertainty methods in the medical domain. Using RNN ensembles and various Bayesian RNNs, we show that population-level metrics, such as AUC-PR, AUC-ROC, log-likelihood, and calibration error, do not capture model uncertainty. Meanwhile, the presence of significant variability in patient-specific predictions and optimal decisions motivates the need for capturing model uncertainty. Understanding the uncertainty for individual patients is an area with clear clinical impact, such as determining when a model decision is likely to be brittle. We further show that RNNs with only Bayesian embeddings can be a more efficient way to capture model uncertainty compared to ensembles, and we analyze how model uncertainty is impacted across individual input features and patient subgroups.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

Citations in the corpus

Underspecification Presents Challenges for Credibility in Modern Machine Learning (2020, methodology): “The experimental protocol we use to probe underspecification is closely related to uncertainty quantification approaches based on deep ensembles (e.g., Lakshminarayanan et al., 2017; Dusenberry et al., 2020).”
Efficient and Scalable Bayesian Neural Nets with Rank-1 Factors (2020, methodology): “Finally, on the MIMIC-III electronic health record (EHR) dataset (Johnson et al., 2016) with LSTMs, rank-1 BNNs outperform deterministic and stochastic baselines from Dusenberry et al. (2019).”
BatchEnsemble: An Alternative Approach to Efficient Ensemble and Lifelong Learning (2020, methodology): “Bayesian neural networks (Hinton & Neal, 1995), which posit a distribution over the weights rather than a point estimate, are often used for model uncertainty (Dusenberry et al., 2019”
Measuring Calibration in Deep Learning (2019, background): “The reliability of a machine learning model’s confidence in its predictions is critical for high risk applications, such as deciding whether to trust a medical diagnosis prediction (Crowson et al., 2016; Jiang et al., 2011; Raghu et al., 2018; Dusenberry et al., 2020).”
A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges (2020, background): “[428] examined the role of model uncertainty techniques in the medical domain.”
Bayesian Layers: A Module for Neural Network Uncertainty (2018, background): “Bayesian neural networks have been to help address important challenges such as indicating model misfit (Dusenberry et al., 2019), generalization to out of distribution examples (Louizos and Welling, 2017), balancing exploration and exploitation in sequential decision-making (Hafner et al., 2019),…”
Deep learning-enabled medical computer vision (2021, background): “95 with permission. the uncertainty 151 around model outputs.”
Second opinion needed: communicating uncertainty in medical machine learning (2021, methodology): “Variational inference methods 36,37 are one popular class of approximations, but impose stricter assumptions about correlations between model parameters than more ﬂ exible methods 4,38 – 42 .”