Artificial Intelligence in Public Health and Epidemiology.

OBJECTIVES: To introduce and summarize current research in the field of Public Health and Epidemiology Informatics.
METHODS: The 2017 literature concerning public health and epidemiology informatics was searched in PubMed and Web of Science, and the returned references were reviewed by the two section editors to select 14 candidate best papers. These papers were then peer-reviewed by external reviewers to provide the editorial team with an enlightened vision to select the best papers.
RESULTS: Among the 843 references retrieved from PubMed and Web of Science, two were finally selected as best papers. The first one analyzes the relationship between the disease, social/mass media, and public emotions to understand public overreaction (leading to a noticeable reduction of social and economic activities) in the context of a nation-wide outbreak of Middle East Respiratory Syndrome (MERS) in Korea in 2015. The second paper concerns a new methodology to de-identify patient notes in electronic health records based on artificial neural networks that outperformed existing methods.
CONCLUSIONS: Surveillance is still a productive topic in public health informatics but other very important topics in Public Health are appearing. For example, the use of artificial intelligence approaches is increasing. Georg Thieme Verlag KG Stuttgart.

Introduction

As quoted in the synopsis of the Public Health and Epidemiology Informatics section of the 2017 IMIA Yearbook 1 , precision public/global health and digital epidemiology are terms that are still in use in 2018 2 3 . The first term is about providing the right intervention to the right population at the right time 2 . The second term is about the use of digital data, especially those that were not collected on purpose, to answer epidemiologic questions 3 . Both refer to the unforeseen opportunities provided by our digital world and new technologies. Although genomics (and more broadly any “-omics”) data continue to contribute, as it is the case for precision medicine, there are many other sources of information that can be used: social networks, internet search engines, cell phone data, electronic health data, and more. The challenge today is to analyze these big data in a meaningful way. One recently improved method that showed very nice success especially in image analysis is deep learning 4 . Applications of this method appear to be only limited by the quantity of information available. Predicting the unplanned readmission at the hospital within 6 months based on electronic health data 5 , de-identifying electronic health records (EHRs) 6 , analyzing social media 7 8 9 are various types of applications relevant in epidemiology and public health. But artificial intelligence covers many other techniques, such as machine learning approaches and statistical learning that offer a panel of methods which usefulness is only limited by pairing them with the right question; the two best papers of this year section are very good examples 6 7 . Naively mining any large dataset will not give immediate answers. Epidemiologic approaches start with clever and appropriate questions, careful collection of relevant data with the most appropriate design, and validation of the results.

Paper Selection

A comprehensive literature search was performed using two bibliographic databases, Pubmed/Medline (from NCBI, National Center for Biotechnology Information), and Web of Science® (from Thomson Reuters). The search was targeted at public health and epidemiology papers that involve computer science or the massive amount of web-generated data. References addressing the topics of the other sections of the Yearbook, such as those related to interoperability between data providers were excluded from our search. The study was performed at the beginning of January 2018, and the search over the year 2017 returned a total of 843 references.

Articles were separately reviewed by the two section editors, and were first classified into three categories: keep, discard, or leave pending. Then, the “keep” and “leave pending” lists of references built by the two section editors were merged, yielding 97 references. The two section editors jointly reviewed the 97 references and drafted a consensual list of 14 candidate best papers. All pre-selected 14 papers were then peer-reviewed by Yearbook editors and external reviewers (at least four reviewers per paper). Two papers were finally selected as best papers ( Table 1 ). A content summary of these selected papers can be found in the appendix of this synopsis. Lamy et al. 10 describe the whole selection process.

Table 1

Best paper selection of articles for the IMIA Yearbook of Medical Informatics 2018 in the section ‘Public Health and Epidemiology Informatics'. The articles are listed in alphabetical order of the first author's surname.

SectionPublic Health and Epidemiology Informatics
▪ Choi S, Lee J, Kang MG, Min H, Chang YS, Yoon S. Large-scale machine learning of media outlets for understanding public reactions to nation-wide viral infection outbreaks. Methods (201 7) 129:50-59.
▪ Dernoncourt F, Lee JY, Uzuner 0, Szolovits P. De-identification of patient notes with recurrent neural networks. J Am Med Inform Assoc (201 7) 24:596-606.

Outlook and Conclusion

As expected in this section of the Yearbook, the use of digital sources for infectious diseases surveillance leads to many research reports 11 . The originality here is the use of data coming from the EHR 12 , Twitter 8 , or the climate data produced by the US National Aeronautics and Space Administration (NASA) with software 13 dedicated to following the occurrence of infectious disease epidemics of either influenza 8 12 or malaria 13 . All these studies demonstrating the feasibility of new approaches for the surveillance of infectious diseases still need to be validated for confirming their predictive accuracy and generaliz-ability. Interestingly, we also found studies reporting the results of the surveillance of non-infectious diseases, e.g., road trafic crashes 14 and elevated blood pressure 15 . Road traffic injuries represent a public health issue in low-income countries 16 . Therefore, improvement of surveillance systems is required. Bonnet et al. 14 have experimented a simple affordable approach based on the city's National Police road crash intervention service equipped with geotracers that geolocalized the crash sites and sent their positions by short message service (SMS) to a surveillance platform developed by using the open-source tool, Ushahidi. This system implemented in partnership with the National Police in the city of Ouagadougou required acceptance by oficers and authorities. In the other study, the authors showed a new validation of the use of EHR for public health purposes. Here they reproduced the seasonability of blood pressure variations (with a peak in summer) based on the data extracted from EHRs 15 .

The other topic covered by several of the papers selected by the review process concerned the analysis of online social media to better understand the attitudes and beliefs toward a given topic, such as vaccination. A social network analysis of Twitter messages (“tweets”) revealed a semantic network for positive, negative, and neutral vaccine sentiment 9 . Beyond this type of analysis, it is fruitful to predict and understand the dynamics of vaccinating behavior. In another paper, Pananos et al. modeled the interaction between vaccination decisions and disease dynamics where one inluences another in a nonlinear feedback loop 17 . They used the theory of critical transitions to derive indicators that may help public health oficials anticipate when resistance to vaccination might develop and intensify. They applied their approach to data from tweets and Google searches around the Disneyland measles outbreak that occurred in 2015 in California 17 . One of the two best papers described below, analyzed the relationship between Middle East Respiratory Syndrome (MERS), mass media, and public emotions during an outbreak in 2015 in Korea 7 .

Digital materials such as tweets are also a potential tool for communication in public health 18 19 with hopefully an improvement of knowledge and attitudes. However, the indicators and the methods to be used for evaluating social media must be adapted to this specific context. Digital tools used in epidemiology need to be validated as any other measure 20 .

Last but not least, it is important to question how interventions and new knowledge generate corresponding changes in public health performance. This is where indicators, measures are needed 21 .