Precision rehab for patients post-stroke through artificial intelligence

The National Institute of Child Health and Human Development awarded $676,469 to the University of Pittsburgh on July 10, 2026, under the Child Health and Human Development Extramural Research program (CFDA 93.865) for development and validation of artificial intelligence–driven rehabilitation prescription tools for stroke patients.

The project develops natural language processing algorithms to extract rehabilitation information from free-text electronic health record data and creates predictive models integrating both structured and unstructured EHR data to prescribe personalized rehabilitation exercises. Specific deliverables include a rehabilitation ontology and ontology-based NLP algorithms (Aim 1), predictive models for tailored rehabilitation recommendations to maximize functional outcomes (Aim 2), and validation of both NLP and predictive models using the ENACT network (Aim 3). The work addresses gaps in current rehabilitation research by enabling clinicians to identify optimal exercise interventions, dosage, and timing for individual stroke patients, thereby enhancing clinical outcomes and reducing lifetime healthcare costs.

Performance occurs in Pittsburgh, Pennsylvania. The award period runs through June 30, 2031. This is a Project Grant, the standard investigator-initiated R01 mechanism for NIH research.

Abstract

Stroke stands as a leading contributor to disability, with an estimated 89.1 million adults worldwide, and 9.4 million adults in the united states. Rehabilitation interventions are effective in mitigating stroke-related disability, greatly enhancing the quality of life, and reducing life-time health care expenses. While standardized rehabilitation protocols are effective in some patients, the wide variety in recovery patterns seen post-stroke make IT extremely difficult to correctly identify the right rehabilitation intervention (i.e., which exercises) at the right dosage (e.g., duration, sets, reps) at the right time for each patient. The integration of rehabilitation activities and outcomes into electronic health records (EHRS) has offered considerable advantages. However, there are notable gaps in the current research on the use of EHRS in rehabilitation research, particularly the difficulty of using free-text EHR data for big data analytics and the lack of predictive modeling that integrates both structured and unstructured EHR data. Our project recognizes this critical GAP and aims to bridge IT by developing and implementing innovative natural language processing (NLP) algorithms to extract rehabilitation information from free-text EHR data, alongside effective predictive models that utilize both structured and unstructured EHR data. This approach is designed to prescribe optimal rehabilitation exercises tailored to individual patients, thereby enhancing clinical outcomes. To address these gaps, this project outlines three specific aims: aim 1: development of a rehabilitation ontology and ontology-based natural language processing (NLP) algorithms. Aim 2: creation of predictive models for personalized rehabilitation recommendations to maximize function outcomes. Aim 3: validation of NLP and predictive models using the enact network. By accomplishing these aims, our project seeks to revolutionize stroke rehabilitation, making IT more responsive to the individual needs of patients. Through the use of advanced AI, we are poised to offer patients post-stroke a more personalized and effective path to recovery. By demonstrating success in diverse healthcare settings in the large national network, we aim to overcome adoption barriers, ensuring scalability and adaptability to various EHR systems and regional healthcare settings.