Multi-Level Provider Network Analysis

Funding: Philip Templeton ’82M (M.D.)
Corporate Collaborator:  X10SYS Database Corporation

Healthcare providers generally work with a small, trusted network of colleagues, forming informal referral networks. Some networks appear to have better population health management outcomes, and identifying these networks can help us learn from their success and improve population health. Analytically, provider networks are generally constructed from insurance claims data, providing a connectivity matrix based on two providers submitting bills for the same patient during a period of time.  The connectivity matrix may be binary (based on any shared patients) or weighted (based on the number of encounters with shared patients), and undirected (does not account for who is the referring provider) and directional (accounting for which provider is referring and which is receiving).  We know, however, that the connections between providers in such networks are multi-level, and not simply based on patient referrals.  Such connections may be geographically constrained (practice locations), constrained by membership to hospital systems and insurance networks, augmented by professional associations and standard social connections (trust, common activities, subspecialty, etc.). The primary goal of this project is to identify and study multi-level provider networks by integrating relationship data from claims, organizational affiliation, geospatial data, laboratory and radiology testing, logistical relations, and shared professional attributes using associative database technology. If successful, this work will result in new methods of identifying provider networks with excellent patient outcomes for common problems such as diabetes, hypertension, and kidney disease.

Patient Journeys To Care in Chronic Kidney Disease

Funding: Philip Templeton ’82M (M.D.)
Corporate Collaborator: X10SYS Database Corporation

Chronic kidney disease affects millions in the United States, and is generally under-recognized and under-treated. Without adequate treatment, chronic kidney disease progresses to end-stage disease, with a need for dialysis or kidney transplantation. Extensive evidence-based guidelines can direct care, and slow the progression of chronic kidney disease. Effective treatment involves a series of partnerships with between patients and providers, with testing and therapies based on kidney function and comorbid medical conditions. Such a series of events, referred to as a “journey” in graph theory, can be modeled using network graphs. How people traverse that network of provider visits, testing, and medications can then be analyzed using network analytics, and linked to patient outcomes. The primary goal of this project is to compare the optimal “journey” for patients with chronic kidney disease with actual patient journeys, and to develop outcomes analytics and metrics of population health with respect to kidney disease. If successful, this work will result in new methods of identifying optimal patient journeys for chronic kidney disease that lead to excellent outcomes.

Intensive Care Unit Readmission Prevention

Patients with serious and life-threatening medical conditions often are admitted to an intensive care unit (ICU) during their hospitalization.  After improvement and when they require a lower level of care, they are transferred to a lower acuity medical or surgical floor.  However, some patients become sicker, and are readmitted to the intensive care unit. Patients who are readmitted to the ICU have longer lengths of stay and a higher risk of death.  Broad risk factors for returning to the ICU include age, comorbid conditions, diagnosis, ICU length of stay, time of transfer to the lower acuity unit, abnormal vital signs, and ICU occupancy level at the time of discharge. However, robust studies using high-dimensional data mining and predictive statistical modeling, are lacking in the literature. In addition, qualitative data such patient and family assessment of ICU discharge readiness, in-hospital family support, unit culture, and patient psychological factors have not been included in previous analyses.    The primary goal of this study is to leverage the knowledge, intuition and insights of patients, families, and the full range of health care providers, along with high dimensional data and clustering methods, to identify and validate  an ICU readmission risk-profile, prospectively test the profile, and implement measures that inclusively involve the patient, family, and entire care team in preventing ICU readmissions.

Population Modeling of Influenza Immunity

Influenza virus infection is an annual public health problem, causing mild to severe illness in a large proportion of the population.  Risk factors for influenza infection include age, immune status, chronic illness, and the degree of immunity against circulating influenza strains.  The primary goal of this project is to use high dimensional data mining, coupled with a novel high dimensional measurement of influenza immunity, to identify and model population immunity to influenza.  If successful, this work will allow us to predict influenza immunity using a data profile, and improve community health.