Development and Validation of a Prediction Model for Longitudinal Fall Risk in Older Adults - ABSTRACT
My goal is to establish myself as an independent biostatistical researcher who develops and validates
prediction models that are critical to identifying older adults at high risk of adverse health outcomes using novel
machine learning methodology. My passion for developing prediction models for health outcomes in older
adults stems from my desire to help people, especially older adults who are vulnerable. I developed a machine
learning method called Binary Mixed Model (BiMM) forest, which is particularly suited for developing prediction
models for repeated measures of outcomes in older adults because it accommodates dynamic fluctuations
over time which are common in aging (e.g., functional status). Environment: Wake Forest School of Medicine
is a nationally-recognized leader in geriatric research and provides an outstanding environment for
accomplishing my goals. Mentors: I have an excellent interdisciplinary mentoring team consisting of experts in
aging, biostatistics and informatics who are dedicated to supporting me in completing the research and training
aims proposed in this K25 grant. Training: I will complete activities aimed to address gaps in my training with
the guidance of my mentoring team. My Training Objectives are to 1) acquire knowledge about gerontology
and geriatrics, 2) deepen my understanding of missing data mechanisms and techniques, 3) learn about the
appropriate use of electronic health record (EHR) data for research purposes, and 4) develop leadership,
networking, communication and grant writing skills. To accomplish these objectives, I will attend conferences,
seminars, journal club, case conferences and writing workshops, observe a geriatric clinician, and complete
coursework. The K25 grant will allow me to extend previous training, which will provide me superior
knowledge, experience and skills for my future as a biostatistical researcher in aging. Research: Early
identification of older adults at-risk of falling is critical so that preventative care plans may be implemented to
improve patient outcomes and reduce burden on the health care system. Most current prediction models
cannot handle repeated measurements over time and have not been used with EHR data. I propose to use my
BiMM forest method to develop a prediction model for fall risk over time. My Research Specific Aims are to 1)
impute (fill in) missing predictor data in Health, Aging, and Body Composition (Health ABC) Study with a novel
machine learning method (BiMM forest); 2) develop a prediction model for identifying older adults at-risk for
falls; and 3) assess the feasibility of using EHR data to externally validate the prediction model. The prediction
model developed in this proposal will aid in identification of at-risk individuals for falls, allowing providers to
intervene early to reduce the burden of falls for patients, caregivers and healthcare systems. Results from this
grant will provide a basis for future R01 submissions to further validate the prediction model using EHR data
from multiple health care centers and evaluate the clinical utility of the model in practice.
