Improving Adherence and Outcomes by Artificial Intelligence-Adapted Text Messages
Improving Adherence and Outcomes by Artificial Intelligence-Adapted Text Messages
Uncontrolled hypertension is a major cause of morbidity and mortality and many patients fail to take their antihypertensive medication as prescribed. The investigators propose to use artificial intelligence (AI) to allow short message service (SMS or text messages) interventions to adapt to patients' adherence needs and substantially improve medication taking. The aims of the study are to: (1) develop AI methods for adaptive decision-making in human-centered environments and demonstrate the feasibility of the resulting AI-enhanced SMS medication adherence intervention, (2) demonstrate that the intervention can "learn" by adapting the SMS message stream according to patients' medication taking over time, and (3) examine potential intervention impact as measured by improvements in medication adherence and systolic blood pressures. The investigators will recruit 100 patients with uncontrolled hypertension and antihypertensive medication non-adherence. Adherence and other covariates will be measured via surveys at baseline, 3- and 6 months; blood pressures will be measured at baseline and 6 months. Participants will be given an electronic pill-bottle adherence monitor. Participants will receive SMS messages designed to motivate antihypertensive medication adherence. Message content and frequency will adapt automatically using AI algorithms designed to automatically optimize expected pill bottle opening. For Aim 1, the first 25 patients will be enrolled to develop and test alternative RL algorithms and fine-tune the system parameters. For Aim 2, the investigators will examine changes in the probability distribution over message-types and compare that distribution with patients' reasons for non-adherence reported at baseline. For Aim 3, the investigators will examine changes in self-reported medication non-adherence and blood pressure and automatically-reported pill bottle openings. This pilot study will establish the feasibility and potential impact of this novel approach to mobile health messaging for self-management support. The results will be used to support an R01 application for a larger and more definitive trial of intervention impacts.
Self-management of chronic conditions involves complex behaviors, and patients vary in their adherence to these behaviors. The focus of this proposal is medication adherence because patients' failure to take their medications as prescribed is a major cause of excess morbidity and mortality and increased health care costs. Studies suggest that 33-50% of patients do not take their medications properly, contributing to nearly 100,000 premature deaths each year and $290 billion in health care costs. Adherence to antihypertensive medications is of particular importance in its own right, and hypertension can serve as an important tracer condition to better understand and improve medication adherence more generally. Uncontrolled hypertension is a major cause of stroke, coronary heart disease, heart failure and mortality, and medication non-adherence is a major cause of uncontrolled hypertension. For example, in a one-year study of ~5,000 hypertensive patients, most patients took their medications only intermittently with half of patients eventually discontinuing their medications against medical advise.
Improving medication adherence requires addressing multiple challenges because patients typically have a variety of reasons for not taking their medication as prescribed, such as beliefs about their disease and its treatment, organizational challenges, and cost barriers. Moreover, as patients' regimens, health status, and social context change over time, adherence support interventions need to adapt, but most services lack the flexibility to do so.
Mobile health (mHealth) services such as patient text messaging or SMS have shown some promise in improving medication adherence. However, since almost all mHealth services are based on simplistic, deterministic protocols, these interventions lack the capacity to meet patients' complex changing needs. As a consequence, these rudimentary systems have demonstrated only modest effects that tend to decrease over time. The investigators propose to apply artificial intelligence (AI) methods, specifically Reinforcement Learning (one type of AI), to develop a model medication adherence system that can automatically adapt SMS communication to improve individual medication taking.
The proposed project is the result of a new multidisciplinary collaboration between UM experts from the College of Pharmacy, College of Engineering, and School of Medicine. Our long-term goal is to improve health outcomes using artificial intelligence (AI) enhanced mobile health tools. The objective in the proposed pilot study is to develop a Reinforcement Learning-based mHealth program focused on medication adherence among patients with poorly controlled hypertension. Our central hypotheses are that a SMS system that uses Reinforcement Learning (RL) will: be acceptable to patients, adapt to hypertension patients' unique adherence-related needs and preferences and changes in these needs over time, and improve medication adherence and blood pressure control. The specific aims are:
The results of this pilot project will include a novel AI/RL technology and evidence regarding its real-world use based on experience with a sample of adults with poorly controlled hypertension. These results will be used to support an R01 application for a larger and more definitive study of the intervention's impact on patients' health and long-term adherence behaviors. Over the longer term, this AI-enhanced mHealth self-management support infrastructure and unprecedented collaboration between investigators in Pharmacy, Medicine, and Computer Science will lay the foundation for a larger program of NIH-funded research using similar AI approaches to addressing behavior change challenges in a large number of health and healthcare problems.
Inclusion Criteria:
Exclusion Criteria: