By Inna Tokarev-Sela, Head of AI at Sisense
As COVID-19 began to sweep the globe, organizations of all shapes, sizes, and missions sought every available tool in the fight to save lives and limit the scope and spread of the pandemic.
One of the key tools at their disposal was data. G-Med is the world’s largest online medical community, featuring over one million verified physicians from more than 120 countries, spanning 100 specialties.
G-Med understands the urgency of putting the right information in the hands of doctors, while Sisense understands the ability of machine learning (ML) to find needles in a haystack, putting game-changing insights in the hands of medical personnel.
When the COVID-19 outbreak began, G-Med sought to improve the analysis applied to the wealth of information in its global forum. Sisense and Amazon Web Services (AWS) teamed up to make a plethora of COVID-19 treatment protocols easily analyzed and searchable using AWS machine learning technologies.
The Sisense analytics and business intelligence (BI) platform empowers analytic builders to deliver insights on complex data, helping users of all kinds achieve improved outcomes.
Sisense is a leader in BI software and an AWS Partner Network (APN) Advanced Technology Partner with the AWS Data & Analytics Competency. Its data analytics platform enables a centralized data strategy that is tightly integrated with AWS services.
In the spirit of helping battle COVID-19, Sisense joined a hackathon called Hackorona to work on enhancing the ability of doctors to quickly find the right COVID-related content within G-Med’s vast medical information repository.
The result is a new app, which Sisense named Medin’Sight. It provides a centralized data repository to help researchers, physicians, and other healthcare workers understand the trends that are emerging in the global fight against COVID-19.
Identifying trends in patient symptoms, treatments, and clinical outcomes is critical to making progress for the patients currently hospitalized, as well as those who will become ill in the days, weeks, and months ahead.
It was a perfect fit to use Sisense to bring together a wide array of medical information to empower healthcare professionals worldwide. Amazon Comprehend Medical is a natural language processing (NLP) service that makes it easy to use machine learning to extract relevant medical information from unstructured text. It was the natural choice to analyze the underlying data.
Pulling Insights from a Vast Pool of Medical Data
The medical community needed a better tool for sharing, analyzing, and aggregating data transparently. A far-ranging, comprehensive knowledge base could empower them to save lives and eventually help quell the pandemic. Sisense identified G-Med as a place where answers could be found.
G-Med offers a huge knowledge base which contributors from around the world are constantly adding new information to. It contains treatment protocols, questions, and answers around a wide array of topics, as well as forums where healthcare professionals can share their expertise.
The content in G-Med’s knowledge base is rich overall, but a deeper analysis was needed to create more layers of insights. When building Medin’Sight, the goal was to create an easy way for users to analyze content and find actionable trends.
The experts that G-Med brings together discuss medical issues via an online forum community, pre-recorded webinars, live 1-on-1 chats, live group chats, virtual video rooms, email alerts, and more. One of the main advantages of the G-Med platform is exclusivity; it employs a thorough screening system to ensure all users are actually physicians. As a result, users know they’re among peers with whom they can exchange trusted, professional information and advice.
During the coronavirus crisis, the G-Med community and its repository have become essential tools leveraged daily by doctors from five continents. Physicians are learning on-the-go as they try valiantly to curtail the impact of the virus.
Conventional communication methods among physicians—such as conference networking and publications—are not suitable for a time like this when physicians need to act as quickly and precisely as possible. G-Med fills this gap, enabling communication, knowledge-sharing, and instantaneous global networking among physicians from every specialty.
How G-Med Helps Medical Professionals at a Critical Time
G-Med’s community is a vibrant nexus for information-sharing among medical professionals around the world. On a daily basis, physicians report cases, images, test results, and more. They’re constantly reaching out to the international community in search of further information about the cases they’re encountering.
G-Med’s repository has become a database of recommended medications and treatments, with the most varied treatments thoroughly discussed in terms of effectiveness, medication dosage, and possible effects. Drugs such as Remdesivir, Hydroxychloroquine, Favipiravir, Ivermectin, and Dipyridamole have been discussed at length via G-Med, with perspectives and results offered by medical professionals from 160 countries since the early days of the pandemic.
Physicians have published the treatment guidelines from their own countries, comparing them to those of their colleagues in addition to those offered by the World Health Organization (WHO). From the first few weeks of the pandemic, physicians were able to engage in essential discussions about how to address the crisis.
Doctors from Brazil, Colombia, Ecuador, Peru, and several other Latin American countries reached out to their European peers for advice in order to build strategies on how to deal with the approaching virus.
Furthermore, medical professionals without access to reliable government information shared their concerns with G-Med colleagues, bringing the community closer together. G-Med has helped physicians expand the limits of the medical community at a critical time.
How the Solution Works
The Sisense team pulled a data sample from the COVID-19 period (a massive sample with thousands of rows) out of G-Med’s database and imported it into the Sisense platform.
The team then ran Amazon Comprehend on the information to determine the sentiment for each sentence. Sentiment is vital in use cases like this because things change rapidly in healthcare, and the COVID-19 pandemic advanced the pace of change to previously unknown levels.
What was true one day might not be true now. Looking for a treatment protocol by a keyword and getting the sentiment spread was like bringing the wisdom of the crowd into the room for timely medical questions.
Once the data sample was analyzed, Sisense also used Amazon Comprehend Medical to extract the tags from the corpus. In parallel, the team also built a BERT-based content search system using the tags extracted earlier.
The AWS services used in this project were leveraged to extract the sentiment and entities and save the results in our Postgres database as follows:
- Add Post – Send the content of the post and store it in the (Postgres) database:
- comprehend.detectSentiment – get the post sentiment and sentiment score.
- comprehendmedical.detectEntitiesV2 – get the relevant entities, tag score, begun and end offsets, and the category.
- Search Posts – send the search term and return the relevant post related to the tags:
- comprehendmedical.detectEntitiesV2 – get the relevant entities.
- Get Posts – send limit, offset, order_by and parent_id parameters and return the relevant posts.
Figure 1 – Architecture for the Medin’Sight app.
The outcome of this series of steps was the Medin’Sight app, a search utility that pulls answers from the massive G-Med dataset based on the entities extracted by Amazon Comprehend. Its dashboard empowers users to slice and dice sentences based on tags and sentiment.
Here’s an example of how this works:
- Healthcare workers can open the Medin’Sight app and see there is positive sentiment around a tag that’s connected to using a specific medicine with the elderly population.
- They can then drill down to the supporting documents.
- As the G-Med user base continues to add more data and additional visualizations and even surveys, the app becomes even more robust and useful as new tags are added and information is associated with existing tags.
- Running on AWS allows Medin’Sight to offer always-on access to analyze entries from thousands of healthcare workers in every corner of the globe.
In Figure 2 below, you can see sentiment analysis of the main keywords discussed by the COVID-19 group on G-Med’s platform.
Figure 2 – Sentiment analysis.
Changing the Game for Frontline Medical Workers
Two key questions facing the medical community in the early weeks of the pandemic offer good examples proving the Medin’Sight app’s capabilities: the effectiveness of hydroxychloroquine in fighting or preventing coronavirus, and the efficacy of mechanical ventilation in acute cases.
In early April 2020, G-Med used the machine learning and NLP analysis described earlier in this post to present insights from the data it had collected using a dashboard provided by Sisense.
Gaining access to analysis shared in a few discussion threads, comments, and posts by physicians, the Sisense team observed that most of the opinions shared expressed negative or neutral sentiment concerning the use of hydroxychloroquine, an antimalarial drug that was touted by some as a possible treatment or preventative for COVID-19.
An artificial intelligence (AI) analysis of the G-Med data showed negative sentiment about the ineffectiveness of the medication, which preceded the findings of scientific studies later published confirming the same.
The efficiency of mechanical ventilation was another topic debated hourly among physicians on G-Med, starting very early in the pandemic’s journey. Doctors across the globe compared statistics from their hospitals and home countries. They engaged in discussions concerning both the mortality rates of mechanical ventilation and improvements seen when ventilators were used with ECMO (extracorporeal membrane oxygenation).
Based on discussions about country-level differences in mortality rates after patients were given mechanical ventilation, Medin’Sight was able to present results to the community that were later confirmed by respected scientific publications.
In Figure 3 below, you can see the mechanical ventilation survival rate derived from physicians’ experiences, shared within the COVID-19 group on G-Med’s platform.
Figure 3 – Mechanical ventilation survival rate.
Using the tools provided by Sisense, G-Med was able to harness the power of medical crowdsourcing to reach important conclusions in considerably less time than it would have taken if their member physicians were working independently.
With the new Medin’Sight app, G-Med is able to deliver actionable information to the public in a timely manner. This is because information comes directly from physicians and is sorted by AI instead of passing through several stages (such as governments and health organization approvals) that may not contribute to the precision of the information, but could slow down the publication process.
Building a Global Alert System for the Next Pandemic
Combining the G-Med community with Sisense and AWS tools, G-Med is planning to implement a global alert system to inform government bodies and health institutions about approaching threats. This will include sharing the most efficient ways of preventing and dealing with them.
Ilan Ben Ezri, CEO of G-Med, is thrilled with the results and impact Medin’Sight has already had on G-Med’s community of physicians.
“In times of global crisis, innovative global solutions that are based on real-world practice data are needed,” says Ben Ezri. “This is exactly what G-Med is doing. During COVID-19, we realized we had the ability to go even further with our crowdsourced data by accurately analyzing, creating new insights, and presenting the information in a way that was never done before.
“We are delighted to have this partnership with Sisense and Amazon that allows us to do this for the benefit of the global medical community,” adds Ben Ezri.
Here are a few examples of search phrases being made on G-Med’s platform, which now return data-driven answers to an either/or question by extracting topics and finding the matching treatment protocols or discussions:
- “Acetaminophen or Naproxen? Which one is better?”
- “Acyclovir and COVID-19: Anybody know about this?”
- “Chloroquine and Lopinavir/Ritonavir”
Consolidating and making this data searchable allows medical professionals to stay abreast of worldwide developments in treating the virus. Most importantly, it helps them find answers that will give their patients a better chance of going home to their families.
Previously, while the G-Med community excelled at asking questions and providing answers, centralizing, aggregating, and making insights easily searchable was beyond their abilities. That’s where the powerful combination of Sisense’s analytical features and Amazon Comprehend Medical’s ability to extract insights from clinical text have empowered healthcare providers to get better answers, more quickly, with less digging and confusion.
Unified in a Global Battle to Defeat COVID-19
The battle against COVID-19 is far from over and will take countless people from around the world using a wide array of skills to overcome it.
Initiatives like Hackorona, companies like G-Med, efforts from governments, healthcare agencies, and even private citizens are all part of a network of support and care stretching across the globe. A concerted effort can free humanity from this pandemic more quickly. Piece by piece, we are building a new world, with data, analytics, and machine learning playing a major role.
Check out Sisense’s COVID relief package and integrate powerful analytics into your project. Explore the world’s most robust medical information-sharing community at G-Med. And dig into the code that powers this crucial project in these GitHub repositories: UI and service.
The content and opinions in this blog are those of the third party author and AWS is not responsible for the content or accuracy of this post.
Sisense – APN Partner Spotlight
Sisense is an AWS Data & Analytics Competency Partner. Its data analytics platform enables a centralized data strategy that is tightly integrated with AWS services such as Amazon Comprehend Medical.
*Already worked with Sisense? Rate this Partner
*To review an APN Partner, you must be an AWS customer that has worked with them directly on a project.