JMIR Publications

ABSTRACT

Background:

Hospitalizations account for almost one-third of $4.1 trillion healthcare cost in the US. A substantial portion of these hospitalizations are readmissions, which led to Hospital Readmissions Reduction Program (HRRP) in 2012.15 HRRP reduces payments to hospitals with excess readmissions. In 2018, more than $700 million was withheld; this is expected to exceed $1 billion by the year 2022.1 More importantly, there is nothing more physically and emotionally taxing for readmitted patients, demoralizing hospital physicians, nurses, and administrators.

Objective:

Given this high uncertainty of home recovery, intelligent monitoring is needed to predict the outcome of discharged patients to reduce readmissions. Therefore, we developed a remote, low-cost, cloud-based machine learning (ML) platform to enable precision health monitoring, which may fundamentally alter the delivery of home healthcare.

Methods:

Our platform consists of wearable, iPhone-synced sensors connected to our cloud-based ML interface to analyze physical activity remotely and predict clinical outcomes. This system was deployed in skilled nursing facilities where we collected over 17,000 person-day data over 2 years, generating a solid training database. We employed these data to train our XGBoost-based ML environment to conduct a clinical trial, “Activity Assessment of Patients Discharged from Hospital (ACT-I Trial, Stanford University Institutional Review Board Approval #53805),” to test the hypothesis that a comprehensive profile of physical activity will predict clinical outcome.

Results:

We achieved precise prediction of the patients’ clinical outcomes with a sensitivity of 87%, specificity of 79%, and accuracy of 85%.

Conclusions:

We present AiCare’s comprehensive technology solution, consisting of wearable sensors, Bluetooth low energy (BLE)-enabled iOS infrastructure, ML algorithm to implement artificial intelligence, and API-enabled web technology, to measure the daily activities of patients. In this study, remote data collection, robust XGBoost AI analysis and reliable prediction of clinical outcome are reported.