Machine learning advancements in healthcare have made data collected through smartphones and wearable devices a vital source of public health and medical insights. While wearable device data helps to monitor, detect, and predict diseases and health conditions, some data owners hesitate to share such sensitive data with companies or researchers due to privacy concerns. Moreover, wearable devices have been recently available as commercial products; thus large, diverse, and representative datasets are not available to most researchers. In this article, we propose an open marketplace where wearable device users securely monetize their wearable device records by sharing data with consumers (e.g., researchers) to make wearable device data more available to healthcare researchers. To secure the data transactions in a privacy-preserving manner, we use a decentralized approach using Blockchain and Non-Fungible Tokens (NFTs). To ensure data originality and integrity with secure validation, our marketplace uses Trusted Execution Environments (TEE) in wearable devices to verify the correctness of health data. The marketplace also allows researchers to train models using Federated Learning with a TEE-backed secure aggregation of data users may not be willing to share. To ensure user participation, we model incentive mechanisms for the Federated Learning-based and anonymized data-sharing approaches using NFTs. We also propose using payment channels and batching to reduce smart contact gas fees and optimize user profits. If widely adopted, we believe that TEE and Blockchain-based incentives will promote the ethical use of machine learning with validated wearable device data in healthcare and improve user participation due to incentives.

No abstract

BACKGROUND Point of Interest (POI) data (e.g., the locations where people are visiting based on their mobile device movement and self-reports) is increasingly being studied in spatial analysis of pressing healthcare issues, such as for assisting with substance use prevention and treatment. However, the task of retrieving accurate healthcare POI information remains complicated, in part, due to the lack of available POI data except from proprietary sources. With the abundance of open-source projects and commercial geographical databases, POI conflation may be a useful method to enrich spatial data attributes and coverage by merging POI records from different sources. OBJECTIVE This study outlines a proposed framework for healthcare POI conflation and spatial enrichment that involves a multi-step process. METHODS This framework includes the following steps: POI data collection from Open-source Location-Based Services (OLBS), geographic and spatial attributes collection, calculating similarity across datasets, POI matching, spatial attributes enrichment, manual labeling, Quality Control (QC), and a deployable enriched database. We tested the viability of our proposed framework on a drug and substance abuse use case in California, USA. RESULTS Based on our findings, our automated approach was able to detect 11,936 unique POIs related to this healthcare tag. Using a proprietary commercial dataset, we found that 38% of their healthcare POIs were substance use-related POI’s (n = 104 number of substance abuse POIs from SafeGraph). In contrast, our framework, which included a much larger number of healthcare POI’s due to our conflation matrix, was composed of 33% substance use-related POI’s (n = 559 number of POIs). CONCLUSIONS We conclude that using our proposed framework can provide a larger number of POIs with similar relevance and spatial attributes as proprietary commercial datasets, allowing this method to be low-cost method for studying substance use-related POIs.