Contextualized ML-based Predictions for Clinical Settings

Many predictive models are developed and embedded into decision support tools, but fail to be implemented or evaluated for safety, acceptance, and integration into clinical practice. We explore user-centered design for a predictive algorithm’s optimal integration into clinical settings. Our goal is to derive insights for an interventional clinical decision support (CDS) tool that is compatible with existing workflows and practices around evidence-based medicine in complex, team-oriented care environments to ensure safety and impact of a model’s predictive potential.

CDS systems’ lack of successful integration in clinical workflows has been attributed to the incompatibility of AI metrics with those for clinical pattern recognition; the absence of consideration for the sociotechnical systems in which they are embedded; the infancy of explainability methods and resulting misalignment with clinicians’ approach to care; among others. We develop a user-centered implementation of a risk score for delayed cerebral ischemia in patients with subarachnoid hemorrhage that avoids these pitfalls by focusing on contextualization of the prediction among existing clinical data. Our research sought to identify a presentation of the risk prediction as part of a contextualized CDS tool that relies on clinician pattern recognition to align with current practice. We aim to balance clinician and developer perceptions of an algorithm so as to not burden clinicians to dissect the ‘black box,’ and facilitate rapid assimilation of knowledge for effective decision making.

Kayla Schiffer grew up in New York. She obtained her BA at Barnard College, where she studied Medical Humanities and Computer Science. During her time at Barnard, her research focused on using patient generated data to support shared decision making with care providers for poorly understood chronic diseases. She then transitioned to industry, working at Veeva Analytics studying contemporary health care trends and their impact on consumer drug patterns. She is currently a second year PhD student in the Department of Biomedical Informatics at Columbia University, researching under Dr. Chunhua Weng. Her research focus is on knowledge discovery of temporal patterns of disease with an emphasis on interpretability in clinical settings. Kayla recently led a workshop at the American Medical Informatics Association annual conference focused on defining principles of social justice in health informatics.

Show me your (people) and I'll show you my (data): Psycholinguistic and computational approaches to studying bilingual language processing

Most individuals use more than one language in their daily lives, emphasizing the importance of understanding bilingual language processing as a unique phenomenon, distinct from monolingual language processing, which has been a major focus of research up to date. One important facet of bilingual language processing are cross language influences (CLI), namely when processing of one language is modulated by knowledge of another language. In this talk, we will present three studies focusing on how bilinguals process words that share form and meaning (cognates) across the two languages. The first study used methods of corpus analyses and demonstrated that the native language of bilinguals using English can be identified based on their use of cognates with their native language (Rabinovich et al., 2018). The second study used psycholinguistic lab methods and found robust CLI from Arabic and/or Hebrew when bilinguals and trilinguals processed their non-native language, English (Elias et al., in prep.; Schreiber, 2021). Finally, the third study is a joint effort using computational tools to investigate a psycholinguistic question: Namely, are effects of CLI attenuated with growing proficiency in the L2 (Native et al., under review)? We will discuss the added value of using complementary research methods, and the benefits of collaboration.

Shuly Wintner is professor of computer science at the University of Haifa, Israel. His research spans various areas of computational linguistics and natural language processing, including formal grammars, morphology, syntax, language resources, translation, and multilingualism. He served as the editor-in-chief of Springer’s Research on Language and Computation, a program co-chair of EACL-2006, and the general chair of EACL-2014. He was among the founders, and twice (6 years) the chair, of ACL SIG Semitic. He is currently the Chair of the EACL.

Anat Prior is an associate professor in the Faculty of Education at the University of Haifa. I study individual differences in language learning, minority students’ literacy performance, foreign language learning, interactions between first and second language systems and domain general vs. specific bases of language processing. Our goal is to better characterize the interactions between two languages (or more) in a single cognitive system, and to identify the underlying mechanisms leading to individual differences in language learning and processing. Ultimately, we aim for our research to foster the development of effective instruction and intervention programs in the domain of foreign language, and programs targeted at minority language students in mainstream education.