Big Data–Driven Health Portraits for Personalized Management in Noncommunicable Diseases: Scoping Review

TL;DR

This review maps how big data health portraits are used for managing noncommunicable diseases, highlighting gaps in data integration and validation.

Contribution

The study introduces a standardized framework using the 3V model to evaluate health portraits for NCD management.

Findings

Only 17.78% of studies met all three 3V criteria (volume, velocity, variety).

Recommender portraits outperformed other types in external validation and 3V criteria.

Most studies used structured data, with limited use of unstructured data and domain-specific attributes.

Abstract

Health portraits powered by big data integrate diverse health-related data into actionable insights, thereby facilitating precise risk prediction and personalized management of noncommunicable diseases (NCDs). Despite their promise, the adoption and application of health portraits remain fragmented, primarily due to the lack of a standardized conceptual and methodological framework necessary to fully harness their capabilities.

This study aimed to systematically map and categorize existing research on health portraits in the context of NCD management, evaluate how big data has been used through the lens of the 3V (volume, velocity, and variety) framework, assess the extent of external validation and comprehensiveness, and identify challenges, emerging opportunities, and future research directions in this field.

A scoping review was conducted following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines and 6-step framework of Levac et al. A comprehensive search was performed in PubMed, Embase, EBSCO, Ovid, Scopus, Web of Science, and Springer Link, focusing on observational and interventional studies using big data, public databases, electronic health record systems, wearables, and sensors for NCD management from January 2014 to July 2024. Data extraction included study characteristics, modeling approaches, and external validation. Analytical synthesis was conducted using keyword analysis, the 3V framework, and visual tools such as scatter plots, heat maps, and radar charts.

A total of 8707 records were identified, and 89 studies were included for full-text analysis. These studies were categorized into 4 types of health portraits: diagnostic, prognostic, monitoring, and recommender. Evaluation based on the 3V framework showed that only 17.78% of studies met all 3 criteria. In terms of volume, structured data were widely used (64.29%-100% depending on portrait type), while unstructured data usage varied significantly (19.05%-93.33%). Regarding velocity, monitoring and recommender portraits showed high reliance on digital interactive data (over 85%). For variety, only 31.11% of studies incorporated all 3 data attributes (natural, domain, and specific attributes). In terms of comprehensiveness, only 30% of studies reported the external validation, and only 10% met both the external validation and 3V criteria, with recommender portraits outperforming the other types.

This study provides a standardized lens through which to evaluate the development and application of health portraits in NCD management. The findings underscore the need for more robust data integration strategies and emphasize the importance of artificial intelligence–enabled approaches. Furthermore, enhancing external validation and addressing ethical and privacy considerations are critical for advancing the implementation of personalized health management solutions.

Figures