Early warning system to detect health trend anomalies

Early warning system to detect health trend anomalies by means of social media, data analysis, and artificial intelligence.

Christian Mendoza-Buenrostro

During pandemics such as the one currently being experienced with COVID-19 and other health care emergencies, governments and their health care systems must be able to react quickly to protect the health of its citizens and the countries’ economies. However, current experience shows that most countries were not fully prepared to handle such pandemic. Hospitals quickly became overcrowded as new cases of infection increased; both human resources as well as personal protective equipment and hospital supplies were not enough to handle the increased patient volume or they were not properly distributed and managed according to needs from specific areas. 

Governments and hospital administrators must have up-to date real-time information about the pandemic evolution to better create policies and strategies to effectively manage the pandemic. The current approach to obtain information about the evolution of the pandemic is primarily reactive. This approach generates a response time that is not ideal in such scenarios. Health care system administrators, policymakers and government officials require support in obtaining relevant and timely information about the pandemic evolution to better identify potential geographical hotspots of new cases, and therefore respond appropriately by managing necessary resources to coordinate an optimal reaction to the trend changes.

The intended audience of this solution are policymakers such as healthcare administrators and governments officials. The solution described will provide actionable intel about health care emergencies trends by regions. Health care administrators and government officials could leverage this intel to better respond to the needs of the pandemic or health care emergency. In our development plan we include constant feedback from our intended audience. The project development is proposed in a stage-gate approach where there are multiple stages of development and in specific stages there is a feedback scenario from health care administrators and policymakers about the analyses results, generated insights and user interfaces. The feedback obtained will help to improve our solution and better adapt to the specific needs of the end users.

Our solution proposal could have a broad impact in society specially during disease outbreaks or pandemics. People continue to use social media platforms when subjected to stay-at-home measures such as the ones implemented by countries during our most recent pandemic experience, and many of the communications relate to the particular emergency situation. By providing policy makers, healthcare system administrators and government officials with a tool to monitor and predict trends by geography in near real-time regarding the healthcare emergency, they will be in a better position to create policies and take decisive measures towards the protection and care of the population. The actions taken with this information could potentially allow them to allocate the resources needed according to the trend changes obtained by the solution. This could potentially help in reducing the infectious cases by providing more clearer information even to the main population. Even more, in a future development stage several dashboards could be obtained from this solution and shared with the main population through means of webpages or social media feeds, this with the intention of providing a better insight into the current healthcare emergency and to avoid crowded places or plan ahead their movements, events, and actions.

In our solution about an early warning system for disease outbreaks, one of our goals is to help healthcare administrators and decision makers to formulate optimized public health policies. Governments can optimize their reaction to attend the needs of the general population during the pandemic or disease outbreak, increasing their capacity to provide healthcare for vulnerable groups as well as the main population. 

The optimized public health policies drive several strategies such as mobility reduction in localized areas; warning messages to the general population using real-time data; and management of healthcare personnel and medical resources.

Future expansion for our solution: We plan to start with a small-scale group test and verify our results during our first year. Then, in our second year and after our first validation stage we will increase our focus group to city-wide level and further verification and adjustments in preparation for further expansion. Lastly, at the third year we plan to focus on a wider area group and move into a regional deployment in Latin America. 

Since our project has Mexico as its team base location, during our first and second year we will be focusing in developing our solution from the city of Monterrey, Nuevo Leon, Mexico. Further growth implementation will scale up to Mexico and other Latin America countries. We will use cloud infrastructure during our scale-up phases in order to obtain the required computing resources.

We will monitor and measure the performance from the solution to detect outbreaks, identification and response time from healthcare system and government officials. A mean time to anomaly detection (mtad) of 1 hour will be set as a key performance indicator (kpi), in comparison to current time to anomaly detection of days. Ratio of real anomaly detections vs false positives higher than 85%. Frequency of data collection of 5 minutes, and time for data readiness of less than 5 min are other of the indicators to be used during the development of the solution.

Financial, technical, cultural, market, and privacy barriers could exist for this project. Some of the API development tools cost a monthly fee and this should be considered during the project phases as well as with the deployed solution. Additional fees include processing costs and equipment maintenance or cloud services costs. A proof-of-concept prototype can be developed without using much financial resources but the final deployed solution will incur in a higher cost when deployed nationwide.

There is also the risk of the algorithms not being able to predict with enough confidence the desired results from the media feeds. Collaboration to test multiple solution ideas is a potential avenue to mitigate this risk.

Cultural and market barriers relate to each other in terms of social media usage, main social media include Twitter and Facebook but if a large portion of the population does not use any of the main social media then there will be no possibility to get the required data for the solution to perform accordingly.

Privacy policies implemented in current social networks and governments could represent another potential source of barriers.

Tecnologico de Monterrey University. At Monterrey, Nuevo Leon, Mexico.

Financial, technical, cultural, and market barrier could exist for this project.

The project has to be sustainable in order to succeed, this is the reason why it is important to have an initial funding for the project. 

Additionally, we believe it is a challenging project with complexity from the AI/ML algorithms to implement and obtain precise and actionable information for the objective of this project. It would be helpful to join together with engineers from some of the Trinity Challenge partners, which could help accelerate the project development by providing insight and domain knowledge from both social media and AI related domains. 

Finally, we are no experts in terms of cultural and market barriers, therefore partnering with international companies with experience in a multi-cultural global setting would be a welcome accelerator in later developments of this project.

This project requires the use of Twitter and Facebook API´s in order to obtain relevant public data from their social media feeds according to the objectives of this project. Therefore, we would like to partner with both organizations. One approach of collaboration could be a small joint team of developers working from our end and theirs to quickly develop a proof of concept prototype of this solution. In addition, Microsoft could be a potential partner, we could use a hybrid approach using their online infrastructure and services for the development and deployment of this project.

On the other hand, another possible partnership during this project would be to collaborate with research groups from NTU Singapore and NUS Singapore. Both Universities are listed as founding members of the Trinity Challenge and they have top-notch research groups in engineering and technology such as Artificial Intelligence, Natural Language Processing, Data Analytics, and more. We are also open for collaboration with other Universities and research groups as well which might be interested in joining efforts for the development of this solution.