Developing a Novel Analytical Toolbox to Tackle Multifaceted Statistical Challenges in Analyzing Post-Fracture Recovery Trajectories in Older Adults with ADRD

University Of Maryland Baltimore, Baltimore MD

Investigators

Linked publications & trials

Abstract

Project Summary/Abstract Older adults diagnosed with ADRD are up to three Ɵmes more likely than cogniƟvely intact older adults to sustain a hip fracture, and paƟents with ADRD have poorer funcƟonal outcomes, greater disability and dependency, and spend more than 50 fewer days at home in the year aŌer fracture. However, this growing populaƟon is highly heterogeneous with some paƟents experiencing very slow to very fast recovery, which precludes proacƟve risk straƟficaƟon, hinders shared decision making, and thwarts opƟmal transiƟonal care support. Given the significant costs and consequences of hip fractures among older adults, improving recovery trajectories for those with ADRD is a crucial naƟonal priority. Unfortunately, clinical characterisƟcs and hospital-level factors associated with longitudinal post-fracture recovery in this populaƟon are poorly understood, hindering the development of effecƟve and personalized transiƟonal care strategies. Moreover, hospitals oŌen obtain access to Medicare data and outcomes on their clinical populaƟons, but how effecƟvely they can use this data for quality improvement is in quesƟon, which reflects a major missed opportunity to both improve and tailor care for older adults, parƟcularly those with ADRD. Untangling mulƟ-level variabiliƟes within the populaƟon of paƟents with ADRD is criƟcal because they could be the target of more individualized caregiving strategies to promote aging in place, facilitate resource allocaƟon among hospitals, and enable the advancement of precision healthcare. To this end, we will develop, validate, and apply novel analyƟcal methods in data science, which include proposing machine- learning assisted high-dimensional regression, computaƟonally efficient individualized dynamic predicƟon, and mulƟ- algorithm-based robust causal inference methods: Aim 1: Develop a novel machine learning-assisted method for idenƟfying unique paƟent characterisƟcs leading to poor longitudinal recovery outcomes in geriatric seƫngs with mulƟ- level structured data. Aim 2: Develop a novel joint modeling approach for mulƟ-level and mulƟ-variate outcomes: uncovering shared mechanisms and facilitaƟng individualized dynamic outcome predicƟon. Aim 3: Develop a new method of ML-algorithm ensemble to idenƟfy causal factors, as potenƟal target for health system-level and pragmaƟc intervenƟons to enhance recovery outcomes. Aim 4: Leverage Medicare data from >20,000 paƟents treated by over 1000 hospitals to understand mulƟlevel variabiliƟes of post-fracture recovery outcomes for older adults living with ADRD. The proposed method can effecƟvely handle high dimensional data, address mulƟple biases due to informaƟve clustering at mulƟple levels (healthcare facility, individual, observaƟon) and truncaƟon by death, and outperform exisƟng methods and lead to unbiased analyses that disentangle mulƟ-level variability of post-fracture outcomes. Significance is enhanced by developing and releasing soŌware (e.g., R packages) to increase the methods’ uptake among the scienƟfic community. The ulƟmate impact is to enable individualized predicƟon for older adults living with ADRD and promote Ɵmely strategies to improve caregiving for detected high risk cohorts.