Activity-Aware Prompting to Improve Medication Adherence in Heart Failure Patients

The long-term objective of this project is to improve human health and impact health care delivery by developing intelligent technologies that aid with health monitoring and intervention. The immediate objective of this project is to design, evaluate and validate machine learning-based software algorithms that recognize daily activities, provide activity-aware medicine reminder interventions and provide insights on intervention timings that yield successful compliance. The investigators hypothesize that many individuals with needs for medicine intervention can be more compliant with their medicine regimen if prompts are provided at the right times and in the right context. The investigators plan to accomplish these objectives by 1) enhancing and validating software algorithms that recognize daily activities and activity transitions, 2) developing and validating activity-aware medicine prompting interventions for mobile devices, and 3) designing technologies to analyze medicine reminder successes and failures. The proposed work will partner real-time methodologies for validation and algorithmic development with smart phone data, utilize novel activity discovery algorithms, and employ activity recognition and prediction algorithms in the development of activity-aware prompting.

The investigators hypothesize that many individuals with needs for medicine intervention can be more compliant with their medicine regimen if prompts are provided at the right times and in the right context. They will validate the hypothesis by designing and evaluating machine learning-based software algorithms that recognize daily activities, provide activity-aware medicine reminder interventions and provide insights on intervention timings that yield successful compliance. The first aim of the project is to expand and validate software algorithms that recognize daily activities and activity transitions with mobile devices. The hypothesis is that daily behavior contexts can be characterized and tracked with minimal user input using machine learning combined with automated activity discovery. In earlier work, the investigators had demonstrated the success of our algorithms in smart homes. In this project, they propose to adapt the techniques for mobile devices. The second aim of the project is to develop activity-sensitive medicine prompting and assess the impact of activity-sensitive prompting on the primary outcome of medication adherence rates and the secondary outcome of quality of life. To this end, this goal can be decomposed into two tasks including (a) developing activity-sensitive prompting; (b) assessing the impact of activity-sensitive prompting on patient outcomes. The investigators will combine an activity prompting interface with activity recognition to deliver prompts in contexts with demonstrated success. Finally, in the third aim, the investigators design machine learning algorithms to analyze medicine reminder success and failure situations. They hypothesize that machine learning techniques can be used to automatically predict prompt compliance by using computer algorithms to learn how to distinguish successful from unsuccessful prompt situations. In their approach, the investigators utilize sensor data to analyze daily behavior and link behavior context with medicine adherence.

Participants receive medication reminders on a smartphone. The reminders are generated through machine learning algorithms that automate the process of medication prompting according to successful medication contexts that occurred in the past.

The Russell's adherence score will be used to measure medication adherence rate. A 3-hour window centered on the prescribed dosing time will be considered. A dose taken within this time window will be given a full score for that dosing time; a dose taken outside the window but within a 6 hour window will be given a half score for that dosing time; and missed doses will receive a score of 0. Each participant will be assigned a score from 0.0 to 1.0 for each day. The scores for each subject will be averaged to obtain weekly adherence rates. The overall adherence rate will be computed by taking an average other the entire study period.

Inclusion Criteria: have a diagnosis of HF and recently hospitalized for HF exacerbation age ≥ 21 years; live independently (not in an institutional setting); and willing to carry the smartphone throughout the day. Exclusion Criteria: any serious co-morbidities (e.g. malignancy, neurological disorder), impaired cognition, inability to understand, read, write, or speak English or Spanish major or uncorrected hearing or vision loss.

We do not expect to share the data for this feasibility study. We, however, consider any requests for data access from other investigators. Those requesting the data will provide a data-sharing agreement to protect the participant's privacy and will receive only de-identified data.