Jaime Lynn Speiser, PhD, MS
Associate Professor of Biostatistics and Data Science
speiser10 at gmail.com
Twitter/X: @DrJSpeiser
About Me
I am a (bio)statistician focused on prediction modeling with applications in medicine. My work involves developing novel machine learning methodology for prediction modeling, providing guidance for best practices for developing prediction models, and collaborating with medical researchers to develop models that inform clinical decision making in the learning health system. I am currently funded by a Career Development Award from the National Institutes of Health (K25) to develop a missing data imputation method for longitudinal data and develop and validate a prediction model for fall risk in older adults.
Shameless Plug
I am always interested in meeting other researchers and students. Please reach out if you'd like to collaborate!
Research
Machine learning prediction modeling
I work in an emerging and competitive area that involves developing novel machine learning methodology for clustered and longitudinal data, which arise when outcomes are correlated within a group or collected repeatedly throughout a study. I am proud to be among the first to publish these powerful and practical models that address challenges that arise with correlated data. Motivated by the longitudinal nature of many cohort studies aimed to assess outcomes over time, I invented a decision tree method called Binary Mixed Model (BiMM) tree and random forest method called BiMM forest that can handle clustered and longitudinal data. Additional work involved incorporating variable selection within the BiMM forest method. This is important because it facilitates removal of superfluous variables in the model, thereby reducing the burden of data collection and improving ease of prediction due to model parsimony. This work resulted in a solo-authored publication in a top bioinformatics journal , as well as a paper that provides practical guidance about choosing optimal random forest variable selection methods for different types of datasets. The latter paper has been cited over 200 times in the first three years since being published and is one of the top cited papers in the journal Expert Systems with Applications in the past three years, as of March 2022.
Collaborative research
Aside from methodology development and practical recommendations for modeling, I enjoy collaborating with scientists and researchers in various medical fields. My role as the lead statistician on a research team involves designing studies, conducting analysis, interpreting results, and writing the statistical methods section for the publications. Early in my career, many of my collaborations focused on acute liver failure and acute liver injury, a rare condition characterized by encephalopathy and severe coagulopathy. I served as lead statistician for studies that evaluated biomarkers: fatty acid binding proteins and regeneration linked microRNA, assessed clinical outcomes over time, and tested a new therapy: molecular adsorbent recirculating system. Aside from collaborations in acute liver failure, studies to which I have contributed include developing a machine learning model for classifying types of mitochondria, analyzing use of contact lenses after surgery, assessing food insecurity in obese individuals, evaluating a structured rounding program at a large healthcare center, and analyzing the association between C-reactive protein, Lipoprotein(a) and cardiovascular disease.
Links
Recent Presentations
Machine learning prediction modeling for longitudinal outcomes in older adults
Seminar Presentation
Machine learning in aging: Who, what, when and how?
Seminar Presentation
Best practices for reviewing papers with machine learning
Seminar Presentation
Machine learning prediction models for mobility limitation over time in older adults: the Health ABC study
Gerontological Society of America 2021 Annual Scientific Meeting
Collaborating to develop prediction models: Advice from a recent project
Women in Statistics and Data Science 2021 Conference
Development of machine learning prediction models: Which random forest variable selection methods are best?
Association for Clinical and Translational Science: Translational Science 2019 Meeting
So you developed a clinical prediction model, now what? Considerations for critical evaluation and implementation
Joint Statistical Meetings 2022
The Unlikely Methodologist: A hodge-podge talk about my research and lessons learned along the way
Seminar talk from WFU 2022