Public Health and Epidemiology Informatics: Recent Research Trends Moving toward Public Health Data Science

Sébastien Cossin, Rodolphe Thiébaut, Section Editors for the Imia Yearbook Section on Public Health and Epidemiology InformaticsYearbook 2020 · DOI

Abstract

The ever-increasing amount of data available offers endless opportunities to develop methods and tools that could assist public health surveillance and intervention belonging to the growing field of public health Data Science. The transition from proofs of concept to real-world applications and adoption by health authorities remains a difficult leap to make.

Introduction

The landscape of public health informatics underwent a definitive shift in 2019, moving away from traditional, often reactive epidemiology toward a more agile and proactive "Public Health Data Science." This evolution is driven by the necessity to process vast, non-traditional data sources—ranging from satellite imagery to global news feeds—to address public health challenges in real-time. While the potential is immense, the field faces a significant "difficult leap" in moving these academic innovations into the hands of health authorities for real-world application.

Methodology and Selection Process

As part of the International Medical Informatics Association (IMIA) Yearbook, researchers conducted a multi-stage literature synthesis to identify the most impactful work of the year. Starting with a pool of 835 references from PubMed, section editors narrowed the selection to 14 candidate "best papers." Following external peer review, two papers were highlighted for their exceptional contribution to the field, representing the cutting edge of computer vision and natural language processing (NLP) in public health.

The 2019 best paper selections for the Public Health and Epidemiology Informatics section, highlighting the work of Bruzelius et al. and Feldman et al.

Case Study 1: Satellite Imagery as a Proxy Census

In many low-income regions, the lack of basic infrastructure like census data makes planning health interventions nearly impossible. Bruzelius et al. addressed this by using computer vision and deep learning. By employing the TensorBox framework on satellite imagery, they were able to identify individual buildings and cluster them into communities.

This method proved highly effective, detecting 75% of registered communities and identifying 167 additional building groupings previously unknown to local health authorities. This "proxy census" allows for the accurate planning of catchment areas in regions where vital registration systems are weak or non-existent.

Case Study 2: NLP for Global Outbreak Surveillance

The second highlighted study, by Feldman et al., focused on the challenge of "signal vs. noise" in global health security. By leveraging the GDELT project—a massive database of global news—and combining it with supervised machine learning across 65 languages, they developed a system for real-time infectious disease monitoring.

The results were staggering: the system achieved an F1 score of 0.87 and correctly detected 94% of outbreaks before the World Health Organization (WHO). Most notably, it provided a mean lead time of 43.4 days earlier than official reports, offering a critical window for early intervention.

Beyond these two studies, several key trends emerged in 2019 that define the current research trajectory:

  • Nowcasting and "Deceptiveness": Researchers are increasingly using search queries to measure disease activity in real-time. To improve these models, the concept of "deceptiveness" was introduced to quantify noise in model features, helping to explain why certain digital signals fail to predict actual outbreaks.
  • Digital Segmentation: Borrowing from targeted advertising, researchers are using NLP and data mining to identify high-risk sub-populations (e.g., specific tobacco users) to deliver tailored prevention messages.
  • The Ethical Consensus Gap: A critical observation is the disparity in oversight. While Electronic Health Record (EHR) research is strictly regulated by Institutional Review Boards (IRBs), social media mining often lacks a global ethical consensus, despite significant privacy risks.

Conclusion and Future Outlook

The shift toward Public Health Data Science is well underway, but technical wins must be matched by practical integration. The "proof of concept trap" remains a hurdle; tools must be integrated into existing health authority workflows to be truly effective. Furthermore, as mobile phone ubiquity grows, the potential for large-scale participatory surveillance increases, provided the field can navigate the pressing needs for data quality and ethical privacy standards.

Summary updated

Abstract

The ever-increasing amount of data available offers endless opportunities to develop methods and tools that could assist public health surveillance and intervention belonging to the growing field of public health Data Science. The transition from proofs of concept to real world applications and adoption by health authorities remains a difficult leap to make.
BibTeX
@inproceedings{cossin2020public,
  author = {Sébastien Cossin and Rodolphe Thiébaut and Section Editors for the IMIA Yearbook Section on Public Health and Epidemiology Informatics},
  title = {{Public Health and Epidemiology Informatics: Recent Research Trends Moving toward Public Health Data Science}},
  booktitle = {Yearbook},
  year = {2020},
  doi = {10.1055/s-0040-1702020},
}
Figure