Search for: All records

We describe HypotheSAEs, a general method to hypothesize interpretable relationships between text data (e.g., headlines) and a target variable (e.g., clicks). HypotheSAEs has three steps: (1) train a sparse autoencoder on text embeddings to produce interpretable features describing the data distribution, (2) select features that predict the target variable, and (3) generate a natural language interpretation of each feature (e.g., mentions being surprised or shocked) using an LLM. Each interpretation serves as a hypothesis about what predicts the target variable. Compared to baselines, our method better identifies reference hypotheses on synthetic datasets (at least +0.06 in F1) and produces more predictive hypotheses on real datasets (~twice as many significant findings), despite requiring 1-2 orders of magnitude less compute than recent LLM-based methods. HypotheSAEs also produces novel discoveries on two well-studied tasks: explaining partisan differences in Congressional speeches and identifying drivers of engagement with online headlines. 

The increased capabilities of generative artificial intelligence (AI) have dramatically expanded its possible use cases in medicine. We provide a comprehensive overview of generative AI use cases for clinicians, patients, clinical trial organizers, researchers, and trainees. We then discuss the many challenges—including maintaining privacy and security, improving transparency and interpretability, upholding equity, and rigorously evaluating models—that must be overcome to realize this potential, as well as the open research directions they give rise to. 

Policy must be informed by, but also facilitate the generation of, scientific evidence 

Abstract The first step towards reducing the pervasive disparities in women’s health is to quantify them. Accurate estimates of therelative prevalenceacross groups—capturing, for example, that a condition affects Black women more frequently than white women—facilitate effective and equitable health policy that prioritizes groups who are disproportionately affected by a condition. However, it is difficult to estimate relative prevalence when a health condition is underreported, as many women’s health conditions are. In this work, we present , a method for accurately estimating the relative prevalence of underreported health conditions which builds upon the literature in positive unlabeled learning. We show that under a commonly made assumption—that the probability of having a health condition given a set of symptoms remains constant across groups—we can recover the relative prevalence, even without restrictive assumptions commonly made in positive unlabeled learning and even if it is impossible to recover the absolute prevalence. We conduct experiments on synthetic and real health data which demonstrate ’s ability to recover the relative prevalence more accurately than do previous methods. We then use to quantify the relative prevalence of intimate partner violence (IPV) in two large emergency department datasets. We find higher prevalences of IPV among patients who are on Medicaid, not legally married, and non-white, and among patients who live in lower-income zip codes or in metropolitan counties. We show that correcting for underreporting is important to accurately quantify these disparities and that failing to do so yields less plausible estimates. Our method is broadly applicable to underreported conditions in women’s health, as well as to gender biases beyond healthcare. 

Despite ethical and historical arguments for removing race from clinical algorithms, the consequences of removal remain unclear. Here, we highlight a largely undiscussed consideration in this debate: varying data quality of input features across race groups. For example, family history of cancer is an essential predictor in cancer risk prediction algorithms but is less reliably documented for Black participants and may therefore be less predictive of cancer outcomes. Using data from the Southern Community Cohort Study, we assessed whether race adjustments could allow risk prediction models to capture varying data quality by race, focusing on colorectal cancer risk prediction. We analyzed 77,836 adults with no history of colorectal cancer at baseline. The predictive value of self-reported family history was greater for White participants than for Black participants. We compared two cancer risk prediction algorithms—a race-blind algorithm which included standard colorectal cancer risk factors but not race, and a race-adjusted algorithm which additionally included race. Relative to the race-blind algorithm, the race-adjusted algorithm improved predictive performance, as measured by goodness of fit in a likelihood ratio test (P-value: <0.001) and area under the receiving operating characteristic curve among Black participants (P-value: 0.006). Because the race-blind algorithm underpredicted risk for Black participants, the race-adjusted algorithm increased the fraction of Black participants among the predicted high-risk group, potentially increasing access to screening. More broadly, this study shows that race adjustments may be beneficial when the data quality of key predictors in clinical algorithms differs by race group.