Hypertension (HTN) is a main risk factor for cardiovascular disease (CVD). It can be managed through use of antihypertensive medications and lifestyle modification (i.e., diet and exercise). Adults who have controlled BP have close to a 50% decreased risk of CVD-related mortality compared to those with uncontrolled BP. However, using current guidelines, over 40% of the 100 million US adults with HTN have controlled BP. Thus, uncontrolled BP remains a significant and largely unchecked public health issue in the United States. Interventions to improve BP control exist (i.e., medication adherence and diet and exercise), but they are mostly complex and resource intensive. If resources to improve HTN control are to be efficiently used, identifying patients who are likely to have persistently uncontrolled BP may provide an opportunity for targeted and tailored intervention and close monitoring. Application of a machine learning algorithm (MLA) to electronic medical record (EMR) data could identify patients likely to have uncontrolled BP. The use and amount of data in EMRs continues to expand and improve in quality. The Temple Health EMR contains extensive demographic, clinical, prescribing, and dispensing data. MLAs are ideally suited to harness this complex and abundant data. MLAs recognize patterns in data that may be difficult to detect by researchers and identify variables that are important for prediction of an outcome. They have been used to predict hospital length of stay, hospital mortality rates, and even detect diabetic retinopathy. The objective of the major study proposed in this K01 application is to first, develop and validate a random forest MLA to predict risk of uncontrolled BP among adults with HTN. Second, identify barriers and facilitators to implementing an MLA-clinical decision support (CDS) tool that predicts uncontrolled BP into clinical settings by conducting key-informant interviews and focus groups with primary care clinicians and patients. Third, Compare clinical management of hypertension with and without a prototype of an MLA-CDS tool through a case-vignette study.
These aims align with my career goal of lowering CVD risk in adults. I will accomplish these aims with the help of an accomplished mentoring team focused on hypertension and antihypertensive medication adherence, machine learning, and qualitative interviewing and CDS-tool development. During the timeframe of this study I will gain experience in (1) MLA development and validation, (2) qualitative interviewing skills to develop CDS tools for clinicians, and (3) leadership skills for principal investigators. These skills will allow me to become an independent investigator focused on identifying people at high risk for chronic disease related to CVD events and developing tools to assist physicians in lowering CVD risk at the population level.
The objective of the major study proposed in this K01 application is to first, develop and validate a random forest machine learning algorithm (MLA) to predict risk of uncontrolled blood pressure among adults with hypertension. Second, identify barriers and facilitators to implementing an MLA-clinical decision support (CDS) tool that predicts uncontrolled blood pressure into clinical settings by conducting key-informant interviews and focus groups with primary care clinicians and patients. Third, compare clinical management of hypertension with and without information from the MLA on whether a patient will have uncontrolled blood pressure through a case-vignette study.