Search for: All records

This work introduces TrialSieve, a novel framework for biomedical information extraction that enhances clinical meta-analysis and drug repurposing. By extending traditional PICO (Patient, Intervention, Comparison, Outcome) methodologies, TrialSieve incorporates hierarchical, treatment group-based graphs, enabling more comprehensive and quantitative comparisons of clinical outcomes. TrialSieve was used to annotate 1609 PubMed abstracts, 170,557 annotations, and 52,638 final spans, incorporating 20 unique annotation categories that capture a diverse range of biomedical entities relevant to systematic reviews and meta-analyses. The performance (accuracy, precision, recall, F1-score) of four natural-language processing (NLP) models (BioLinkBERT, BioBERT, KRISSBERT, PubMedBERT) and the large language model (LLM), GPT-4o, was evaluated using the human-annotated TrialSieve dataset. BioLinkBERT had the best accuracy (0.875) and recall (0.679) for biomedical entity labeling, whereas PubMedBERT had the best precision (0.614) and F1-score (0.639). Error analysis showed that NLP models trained on noisy, human-annotated data can match or, in most cases, surpass human performance. This finding highlights the feasibility of fully automating biomedical information extraction, even when relying on imperfectly annotated datasets. An annotator user study (n = 39) revealed significant (p < 0.05) gains in efficiency and human annotation accuracy with the unique TrialSieve tree-based annotation approach. In summary, TrialSieve provides a foundation to improve automated biomedical information extraction for frontend clinical research. 

Ensembles of decision trees perform well on many problems, but are not interpretable. In contrast to existing approaches in interpretability that focus on explaining relationships between features and predictions, we propose an alternative approach to interpret tree ensemble classifiers by surfacing representative points for each class -- prototypes. We introduce a new distance for Gradient Boosted Tree models, and propose new, adaptive prototype selection methods with theoretical guarantees, with the flexibility to choose a different number of prototypes in each class. We demonstrate our methods on random forests and gradient boosted trees, showing that the prototypes can perform as well as or even better than the original tree ensemble when used as a nearest-prototype classifier. In a user study, humans were better at predicting the output of a tree ensemble classifier when using prototypes than when using Shapley values, a popular feature attribution method. Hence, prototypes present a viable alternative to feature-based explanations for tree ensembles. 

Models which estimate main effects of individual variables alongside interaction effects have an identifiability challenge: effects can be freely moved between main effects and interaction effects without changing the model prediction. This is a critical problem for interpretability because it permits “contradictory" models to represent the same function. To solve this problem, we propose pure interaction effects: variance in the outcome which cannot be represented by any subset of features. This definition has an equivalence with the Functional ANOVA decomposition. To compute this decomposition, we present a fast, exact algorithm that transforms any piecewise-constant function (such as a tree-based model) into a purified, canonical representation. We apply this algorithm to Generalized Additive Models with interactions trained on several datasets and show large disparity, including contradictions, between the apparent and the purified effects. These results underscore the need to specify data distributions and ensure identifiability before interpreting model parameters. 

Black-box risk scoring models permeate our lives, yet are typically proprietary or opaque. We propose Distill-and-Compare, an approach to audit such models without probing the black-box model API or pre-defining features to audit. To gain insight into black-box models, we treat them as teachers, training transparent student models to mimic the risk scores assigned by the black-box models. We compare the mimic model trained with distillation to a second, un-distilled transparent model trained on ground truth outcomes, and use differences between the two models to gain insight into the black-box model. We demonstrate the approach on four data sets: COMPAS, Stop-and-Frisk, Chicago Police, and Lending Club. We also propose a statistical test to determine if a data set is missing key features used to train the black-box model. Our test finds that the ProPublica data is likely missing key feature(s) used in COMPAS.