Authors: Celeste De Marco and Verónica Xhardez
Wednesday, 29 March 2023
ARPHAI was created with the objective of developing Artificial Intelligence (AI) and Data Science-based pilot tools that, applied to electronic health records (EHR), contribute to the early detection of epidemic outbreaks and support preventive public health decision making in Argentina. To this end, ARPHAI prioritized three lines of research and development:
i) Predictive Models
In the first stage of the project, agent-based models (ABM) were used to predict the behavior of variables such as UTI bed occupancy, mortality, symptomatic and asymptomatic cases of COVID-19 in the city of La Rioja. The model integrated a large amount of local public data such as sociodemographic, epidemic surveillance, climate, and mobility, among others, providing good predictions. When analyzing the scalability potential, we deduced that this type of model could be replicated in other cities but did not present the necessary scalability using EHR data. Indeed, we noticed the limitations associated with ABM models: the complexity of calibrating and validating the required local data and the need for parallel computing tools (computer clusters) to process the resulting code in reasonable times.
In a second stage, a new team of researchers was recruited with the challenge of developing a scalable predictive model, in other words, one that uses more EHR data and requires less processing capacity. This complex challenge involved numerous difficulties associated with working with real health data that is "far from perfect". In fact, as a result of applying various types of mathematical and scientific models, several discussions emerged. Among them, how does the phase of the pandemic affect the quality, quantity and availability of critical EHR data, and how to address the challenge of adding details to enhance a model's realism, which significantly impacts its performance and complexity.
ii) Computable phenotypes
The first proof of concept consisted in the automatic identification of a set of COVID-19 phenotypes (suspected, confirmed, negative) using free text (unstructured data ) and EHR codes (IC-10). Eventually, constant changes in the definition of confirmed COVID-19 cases (based on the Argentinian official epidemiological protocol) forced the proof of concept to evolve into another proof of concept exploring the feasibility of automatically detecting symptoms and syndromes using AI in EHR data. Our researchers worked collaboratively with medical experts to identify entities of interest (symptoms, signs, pathologies and terms) in the datasets used and to define the criteria for manual annotation, automatic annotation with rules, and automatic annotation with automatic learning. Based on these annotations, computational rules were established and algorithms were evaluated and trained for both proofs of concept, obtaining very promising results.
Indeed, it is established that in a short time it is possible to achieve a prototype with acceptable results to contribute to the automatic detection of phenotypes and symptoms using EHR data. These results demonstrate the enormous potential of applying AI in EHR to obtain valuable information in order to improve epidemiological management in emergency scenarios.
iii) Epidemiological management dashboards (COVID-19), dengue and syndromic surveillance)
This co-development project in collaboration with public health agencies at national and provincial level has been carried out in a difficult context, marked by the pandemic and successive political-institutional crises. This context has generated ups and downs, difficulties and setbacks that have left us with different lessons learned. We learned of the importance of end-user involvement, as well as the commitment of internal technical teams to ensure their dedication and financial support.
"Health data is not just any type of data: it is essential to evaluate each use scenario with the necessary complexity to safeguard sensitive information and preserve human rights. At ARPHAI we generate some instruments based on our experience in the project that can help other initiatives." - Dra. Sabrina Lopéz, ARPHAI Researcher.
We also learned that it is essential to plan and document the transfer of the research-implementation experience in collaboration with the local technical management teams so that the effort can be sustained independent of institutional ups and downs. Finally, we learned it is important to establish team leaders capable of understanding both the logic and practices of research teams, as well as the times and needs required by public management in order to translate demands and maintain the interest and commitment of both parties.
Last but not least, through the research lines of bias, gender and responsible use of data, ARPHAI has made innovative contributions on gender issues (implementation of the Argentine Gender Identity Law in the Historia de Salud Integrada and recommendations to avoid cissexism and Trans pathologization in Snomed CT medical standard), which supports informed decision making in the public sector. Additionally, this research line has contributed to laying the foundations on how to responsibly manage health data for secondary use in research, and -from ARPHAI’s research experience- participated in the debate and offered recommendations on the public consultant for updating the Personal Data Protection Law in Argentina.
“The institutions and research teams involved in the ARPHAI project have created networks and gathered valuable experience that will strengthen and propel future interdisciplinary projects, fostering research activities that contribute to evidence-based public health decision making.” - Celeste De Marco, Technical Coordinator, ARPHAI Project
In conclusion, over two years of work as part of the Global South AI4COVID Program in collaboration with the different research and development teams involving researchers, technical teams and policy makers from more than 20 institutions in Argentina, the project has laid important foundations for future interdisciplinary projects that require interaction between management and research in the field of health.Tags: COVID-19, Women, Policy Makers