SBIR-STTR Award

Hypertension (HTN) affects 73 million Americans, with an annual hospitalization cost of $113 billion/year. Clinical trial outcomes of optima4BP (O4BP), a clinical reasoning artificial intelligence (AI) system, determined that the success/failure of past anti-HTN medication treatments can be utilized to improve O4BP predictive performance. Our objective for this Phase I proposal is to investigate the feasibility of building a past treatment success/failure ?memory? function into O4BP. O4BP, an innovation of Optima Integrated Health, is functional at UC San Francisco Medical Center (UCSF MC), the clinical collaborating partner. O4BP AI safely optimizes HTN medication treatment independent of in-office visits. It collects and analyzes current patient health information from multiple sources to determine if a medication optimization is required. O4BP then alerts the physician of a need for a medication change by providing a clinically supported treatment recommendation action. In this Phase I, we propose the development and validation of O4BP's ?memory? by using the principles of Instance Based Learning (IBL). Known for its flexibility and intuitive logic, the IBL theory will be adapted to enhance the predictive treatment efficacy power of O4BP. It will lead to incorporating the patient response (success/failure) from past treatments to rank the efficacy of candidate treatments identified by O4BP for HTN medication optimization. Clinical resources [e.g., Electronic Health Record (EHR)] provide incomplete patient information, insufficient to build the ?memory? functionality using retrospective datasets. Therefore, in Aim 1, a clinical trial (n=50) will be conducted to collect complete patient datasets that will be used to build the ?memory? function: (a) BP data from remotely monitored BP; (b) EHR updates to the patient's profile since last cycle data dump; and (c) personalized online surveys for patient reporting of medication adherence, side- effects (SEs), and cardiovascular symptoms. Aim 2 will utilize the datasets to develop an efficacy proximity map 2x/month of candidate treatments to past treatments using IBL theory. The proximity map will be based on scored past treatment(s) efficacy, defined as a combination of success in BP lowering, reducing of SEs, and targeting minimum number of drugs with maximum BP lowering power. The treatment recommendation alert will be sent to the physician, for consideration. This study design allows immediate clinical adjudication of the IBL ?memory? through physician's choice to implement/reject the treatment action alert, made available 2x/month for each patient. ?Memory? enhanced O4BP is expected to lead to a systolic BP (SBP) reduction ?10mmHg by the end of the scheduled 12 months treatment optimization period. Successful development and validation will position the ?memory? enhanced O4BP for deployment at UCSF MC that serves approximately 20,000 patients with an SBP >160mmHg, poorly controlled, and at high risk for primary or recurrent stroke, heart failure or myocardial infarction. Our long-term objective is to transform the reactive and punctuated nature of HTN medication treatment management into a proactive and ongoing component of patient care.

Project Terms:
Accounting; Address; adjudication; Adoption; Adverse effects; Affect; American; Artificial Intelligence; Automobile Driving; base; Blood Pressure; Blood Pressure Monitors; blood pressure reduction; Cardiology; cardiovascular risk factor; Cardiovascular system; Cessation of life; Clinical; clinical decision support; clinical efficacy; Clinical Trials; Collection; college; commercial application; Confidential Information; cost; Data; data exchange; Data Set; design; Development; Diagnosis; Effectiveness; efficacy evaluation; Electronic Health Record; experience; Failure; flexibility; follow-up; Goals; Health; Health Care Costs; Health Insurance Portability and Accountability Act; Heart failure; high risk; high risk population; Hospital Costs; Hospitalization; Hypertension; hypertension treatment; improved; individualized medicine; innovation; Institutional Review Boards; Intelligence; Intuition; Lead; Learning; learning strategy; Life Style; Logic; Machine Learning; Maps; Medical center; medication compliance; Medication Management; Medicine; Memory; Myocardial Infarction; Nature; novel; Office Visits; Outcome; Patient Care; patient oriented; patient response; Patients; Performance; Persons; Pharmaceutical Preparations; Phase; Physicians; Positioning Attribute; prediction algorithm; predictive modeling; profiles in patients; Protocols documentation; Quality of Care; Recommendation; Recurrence; Reporting; Research Design; Resources; response; Risk; Risk Estimate; San Francisco; Schedule; Series; Site; Source; Stroke; stroke incidence; success; Surveys; Symptoms; System; system architecture; Testing; theories; Time; Titrations; Training; Treatment Efficacy; treatment optimization; Update; Validation; vector;

Need. In the US, 40 million patients with hypertension (HTN) have their blood pressure (BP) uncontrolled. BP above clinical Target even for a few months increases the risk for stroke (35-40%), heart failure (HF) (up to 64%), myocardial infarction (MI) (15-25%). Physician-nurse-pharmacist resource-intensive demonstrations in achieving & maintaining BP Target have shown promising results, but their real-life deployment was found unsustainable long-term. As a result, a process-standardized and sustainable solution is acutely needed. Solution. In response to this need, Optima Integrated Health developed optima4BP 1.0. It is a first-in-class artificial intelligence (AI) that simulates the process of clinical reasoning undertaken by the treating physician in optimizing the anti-HTN treatment towards BP Target. Just like the physician, optima4BP 1.0 cannot determine upfront the needed Optimal Treatment (OT) to achieve & maintain BP Target for 1-2 years. PROTECT [optima4BP 2.0: prediction of Optimal Treatment and route to achieve and maintain BP Target] proposes to establish upfront the personalized OT. The OT can then be used to select the shortest and safest treatment modification route needed to achieve & maintain BP Target. Phase II Goal. Build optima4BP 2.0. Phase I. Phase I Prior Work demonstrated that k-Nearest Neighbor (kNN), an AI model, can predict with ? 80% confidence the correct anti-HTN treatment, when compared to physician decision. Phase II. optima4BP 2.0 will predict the Optimal Treatment and route to achieve & maintain BP Target. Optimal Treatment data-mining source. PROTECT will use the SPRINT (Systolic Blood Pressure Intervention Trial, 2015) and ACCORD (Action to Control Cardiovascular Risk in Diabetes, 2010) clinical trial data. They represent the foundation of the most current anti-HTN treatment management national guidelines. Aim 1. Build kNN. Hypothesis. kNN can predict the proximity (clinical relevance) of a patient to an Optimal Treatment (OT). Milestone. Achieve ? 90% accuracy of prediction to physician decision. Phase I Data Preparation protocol will be applied to the SPRINT & ACCORD data. Then, the kNN Ensemble Learning function will be built to select the Optimal Treatment with the highest demonstrated efficacy by comparing the choice from 3 computational approaches developed and tested during Phase I. Aim 2. Build the Optimal Treatment Route (OTR). Hypothesis. Knowing the Current and Optimal Treatment (OT), an OTR can be built. Milestone. Safest Route: Achieve 100% exclusion of treatments that led to an adverse event in similar patient populations. Shortest Route: Achieve ?30% reduction in number of treatment changes compared to physician route. The OTR will be built by comparing at each Step on the Route how similar each Candidate Treatment is to the OT through a computed similarity assessment. optima4BP 2.0 aims to establish a process-standardized & sustainable solution with the goal of reducing the incidence of stroke, HF, MI and death resulting from uncontrolled hypertension.

Public Health Relevance Statement:
In the US, 40 million patients with hypertension have their blood pressure uncontrolled. We propose to build and test optima4BP 2.0 (Project Name: PROTECT) that will predict upfront the personalized optimal treatment and shortest treatment route required by a patient to achieve and maintain blood pressure target. Achieving and maintaining blood pressure target will reduce the incidence of adverse events [stroke (35-40%), heart failure (up to 64%), myocardial infarction (15-25%)] or death resulting from uncontrolled hypertension.

Project Terms:
Acute; Adverse drug event; Adverse event; Antihypertensive Agents; Artificial Intelligence; base; Blood Pressure; blood pressure intervention; California; cardiovascular risk factor; Cessation of life; Clinical; clinical care; Clinical Trials; clinically relevant; commercial application; comparative efficacy; Confidential Information; Data; data mining; data quality; Development; Diabetes Mellitus; Diagnosis; Drug Combinations; Enrollment; Euclidean Space; Exclusion; Foundations; Gender; Generations; Goals; Guidelines; Health; Heart failure; Hypertension; hypertension treatment; Incidence; Individual; innovation; Intervention; Intervention Trial; Learning; Life; Lipids; Logic; Medical center; Mind; Modeling; Modification; Myocardial Infarction; Names; Nature; Nurses; optimal treatments; Outcome; patient population; Patients; Performance; Pharmaceutical Preparations; Pharmacists; Pharmacological Treatment; Pharmacotherapy; Phase; Physicians; Preparation; prevent; Process; processing speed; Protocols documentation; Resources; response; Route; Safety; San Francisco; Seminal; Source; Standardization; Stroke; stroke incidence; stroke risk; success; Testing; Time; Training; Universities; Visit; Women's Health; Women's Role; Work