Projects

Introduction

This initiative addresses the critical need for cybersecurity education in healthcare settings by creating comprehensive training materials tailored specifically for health professionals. The project focuses on bridging the gap between theoretical cybersecurity concepts and their practical applications in real-world healthcare environments.

Introduction

The DETERMINE project will integrate and curate broad social drivers of health across domains and levels to represent holistic contextual health information and explore responsible AI models to uncover intersectional interactions between social and clinical factors of diabetes risk. The team will ensure the model's generalizability across populations by validating it in different healthcare institutions.

Overall, the DETERMINE project will develop an AI-powered multivariable risk prediction model that integrates social, demographic, and clinical factors to deliver accurate, fair, generalizable, and interpretable T2D risk predictions.

Introduction

The team will leverage AI-supported analytics and geospatial modeling to identify ZIP codes with disproportionately high rates of amputations. By integrating clinical, demographic, and geographic data across multiple states, the project will uncover patterns of risk, delayed care, and preventable disease progression. We will identify ZIP codes where earlier intervention could significantly reduce preventable amputations. The project's insights will inform targeted, place-based outreach and prevention strategies for communities at most at risk. Overall, the project aims to reduce preventable amputations, improve health outcomes, and develop a scalable, evidence-based model that can be applied across diverse health settings and regions to advance equitable vascular care.

Introduction

Childhood sexual abuse has serious lifelong impacts on health, wellbeing, and development - it's a devastating adverse childhood experience. While much research has focused on identifying CSA and treatment after harm occurs, far less is known about why abuse recurs and how systems can intervene earlier to prevent recurrent CSA. This project seeks to address that gap by examining the multilevel drivers of recurrent CSA and translating those findings into targeted, system-embedded prevention strategies within the child welfare system. Using a unique, linked longitudinal dataset of child welfare records and law-enforcement investigations in Philadelphia, the team will apply advanced analytic methods and technologies such as natural language processing to extract findings from case narratives and geospatial analyses to incorporate neighborhood context. Quantitative findings are further interpreted through qualitative interviews with practitioners, caregivers, and survivors, ensuring that technology-enabled insights translate into practical, trauma-informed, and system-ready prevention approaches.

Introduction

Cardiovascular disease (CVD) risk is shaped not only by clinical factors, but also by social and structural conditions that influence access to CVD care and prevention. However, existing risk prediction models fail to account for these factors, limiting their effectiveness for populations facing health inequities. The AI2Equity project addresses this gap. We are developing an AI-enhanced, equitable CVD risk model that integrates clinical data with multi-level social drivers of health. Using advanced machine learning methods and diverse healthcare datasets, the AI2Equity project seeks to improve CVD risk prediction and inform targeted interventions that reduce disparities in CVD outcomes.

Introduction

Temple's ALS Center of Hope runs clinical trials but has found that racial and ethnic minority populations are underrepresented. This project is addressing this gap by tackling three key barriers: the high opportunity costs of participation, delayed referrals among individuals from racial and ethnic minority groups, and misperceptions and mistrust about research among patients and community members. To address these barriers, the project integrates trial visits with clinic visits; provides support for parking, transportation, and meals; educates healthcare providers on bias, early diagnosis, and timely referral; and educates community members and people with ALS about ALS, clinical trials, and participant protections.

Project Overview

This project addresses a critical challenge in U.S. healthcare: ensuring that patients with chronic conditions like hypertension and type 2 diabetes receive consistent preventive care. Despite the availability of effective treatments, many patients struggle to maintain regular healthcare visits, leading to increased emergency department use, complications, and healthcare costs.

Research Challenge

Understanding why some patients adhere to recommended care visits while others do not, and why some resort to emergency departments for conditions manageable through routine care, requires analyzing complex interactions between patient characteristics, social determinants of health, healthcare access factors, and provider-level influences. Traditional approaches often examine these factors in isolation, missing important patterns that emerge from comprehensive data integration.

Our Approach

Leveraging data from the HealthShare Exchange, a regional health information exchange serving multiple healthcare systems, this research examines patient- and provider-level factors associated with preventive care visit adherence and emergency department utilization among adults with hypertension, type 2 diabetes, or both conditions. The project employs advanced machine learning and artificial intelligence methods to develop predictive models that can identify patients at risk for poor care adherence or inappropriate ED use.

Key Research Questions

• What patient-level factors (demographics, comorbidities, social determinants) influence preventive care visit adherence?
• How do provider-level and healthcare system factors affect care patterns?
• Can machine learning models accurately predict which patients are at highest risk for care non-adherence or ED overutilization?
• What actionable insights can inform targeted interventions by health systems and insurers?

Impact & Significance

This work aims to provide health systems and insurers with evidence-based strategies for improving maintenance care for chronic diseases. By identifying high-risk patients and understanding modifiable factors that influence care patterns, healthcare organizations can implement targeted interventions to reduce morbidity, mortality, and costs while promoting health equity. Findings will be disseminated to researchers, policymakers, and healthcare stakeholders to inform practice and policy decisions at federal, state, and local levels.

Related Papers

Data Preprocessing and Integration for Healthcare Utilization Analysis in Chronic Disease Management

Focus: Methodological Framework for Health Information Exchange Data Preparation

This paper presents a comprehensive methodological approach for preprocessing and integrating complex health information exchange (HIE) data to support healthcare utilization research in patients with hypertension and type 2 diabetes. The work addresses critical challenges in working with real-world clinical data, including data fragmentation across multiple healthcare systems, missing values, inconsistent coding practices, and the integration of diverse data sources including clinical encounters, laboratory results, diagnoses, medications, and social determinants of health.

Key Contributions:

• Systematic framework for HIE data integration and harmonization across multiple healthcare systems
• Methods for defining and measuring preventive care adherence in chronic disease populations
• Approaches for classifying and characterizing emergency department utilization
• Integration of individual clinical data with neighborhood-level social determinants of health
• Quality assessment procedures for complex, multi-source healthcare data
• Reproducible preprocessing pipeline applicable to other chronic disease utilization studies

Machine Learning Prediction of Healthcare Utilization Patterns in Type 2 Diabetes and Hypertension

Focus: AI-Based Models for Predicting Visit Adherence and Emergency Department Use

This paper develops and evaluates machine learning models to predict preventive care visit adherence and emergency department utilization among adults with type 2 diabetes and hypertension. Leveraging health information exchange data encompassing patient demographics, clinical characteristics, comorbidities, social determinants of health, and healthcare utilization history, the research compares multiple machine learning algorithms to identify patients at highest risk for poor care engagement or inappropriate emergency care use.

Key Contributions:

• Development of machine learning models for predicting visit non-adherence in diabetes and hypertension care
• Predictive models for emergency department use among patients with chronic conditions
• Comparison of multiple ML algorithms for healthcare utilization prediction
• Feature importance analysis identifying key drivers of care patterns
• Clinical validation of AI-based risk stratification approaches
• Practical framework for implementing predictive models in population health management