A Personalized Preventive Care Recommendation System by Integrating Guidelines with the EHR Data

This project addresses a critical gap in current electronic health record (EHR) clinical decision support (CDS) for preventive care, which typically applies broad, age- and gender-based criteria and fixed screening intervals. The project responds to the public health imperative that prevention must play a larger role in improving population health and containing rising healthcare costs: the Centers for Disease Control and Prevention report that a large share of healthcare expenditures goes to people with chronic and mental health conditions, emphasizing the potential value of more effective preventive interventions. The primary objective of the work is to develop a system that generates personalized preventive care recommendations by combining information automatically extracted from preventive care guidelines with patient-specific risk factors identified in EHR data.

The research plan centers on several technical and evaluation components. First, the team will implement an EHR component-based data interchange structure to ensure consistent analysis of extracted guideline and patient information. Second, the project will develop methods to automatically extract relevant recommendations and rationale from clinical preventive guidelines. Third, innovative natural language processing (NLP) and deep learning algorithms will be applied to structured and unstructured EHR data to extract risk factors that are often not captured in discrete fields, for example, social behaviors, family history, symptoms, and other determinants documented in narrative clinical notes. Notably, social and behavioral determinants of health are increasingly recognized as key contributors to disease risk, disability and mortality, yet they are rarely systematically extracted or linked to preventive care recommendations. The proposed NLP approaches aim to bridge that gap by surfacing these underutilized data elements and incorporating them into personalized CDS.

A further emphasis of the project is user-centered evaluation: the system's efficiency, accuracy, and usability will be assessed through studies involving both healthcare providers and patients. The personalized recommendations generated by the system will include not only actionable suggestions but also explicit rationales grounded in the patient's EHR data and the source guideline content, supporting transparency and clinician and patient trust. The project will utilize the Indiana Network for Patient Care (INPC), a statewide clinical data warehouse, providing access to rich EHR data to support method development and evaluation. Long-term aims include automating the integration of multiple preventive care guidelines with EHR data, enhancing patient engagement in preventive care, reducing healthcare costs, and improving population health. The investigators propose that the innovative NLP and deep learning methods developed could be extended to other narrative guidelines and EHR datasets, enabling broader application of personalized guideline-based recommendations across healthcare settings.