Patient care is increasingly guided by evidence based practice guidelines, which are defined as "systematically developed statements to assist practitioner and patient decisions about appropriate health care for specific clinical circumstances." These guidelines are viewed as the key to improving patient care and reducing costs, but current guidelines tend to be specific to individual diseases, and rarely consider all of the relevant details of a patient's condition, such as age, gender, and ethnic background, as well as other diseases from which patients suffer. By addressing the challenge of personalized evidence based medicine, the research in this project will positively impact patients suffering from multiple chronic conditions, which is becoming the norm in the aging US population. To that end, this project will develop new clinical modeling techniques that can use the data available in electronic health records (EHRs) to improve the personalization of these guidelines. More specifically, the ultimate goal of this project is to generate more personalized guidelines that can be implemented in clinical decision support systems and used by physicians and others for the comprehensive treatment of patients with multiple chronic conditions.
To address the challenge of personalized care guidelines to handle multiple chronic conditions, this project develops a modeling framework, Group-Specific Learning (GSL), with the ability to enhance clinical modeling by making models increasingly personalized without rendering them excessively specific. In particular, the GSL modeling paradigm is applied to enhance four modeling techniques commonly used in health sciences research: survival analysis, causal analysis via propensity scoring, competing risk models and multi-state models. This work focuses on type-II diabetes mellitus (T2DM), its precursor, pre-diabetes, its comorbidities (hypertension, obesity, hyperlipidemia), and its consequences (chronic kidney disease, renal failure and the various cardiac and vascular complications). Diabetes has a number of interrelated comorbidities and severe complications, but evidence-based guidelines for the treatment of these conditions treat these conditions in isolation. To address this limitation, this project develops a suite of analytics techniques that can take the substantial heterogeneity that exists in the diabetic population into account in order to measure the effect of existing evidence-based guideline elements (interventions) in terms of risk of progression to diabetic complications. These guideline elements can then be compiled into guidelines, thus allowing for the systemic and comprehensive treatment of the population with heterogeneity.
