In response to Explain why you selected this stage of development for your solution.
"We have also engaged with CBC testing laboratories in India, Hong Kong and Singapore to initiate preservation of CBC data, with the aim of on-boarding them in 2021."
LMICs account for more than 80% of the world's population. India is the only country with planned CBC testing in 2021 (what about future years?). It is crucial that the algorithm is trained and able to profile accurately the blood of LMIC population, especially in zones with higher epidemiologic risks to be an efficient tool.
I completely agree with you that we would love to bring in the LMICs as soon as possible, but our team are also realistic that implementing this global solution requires both algorithm development, cloud infrastructure setup and establishing local data collection at sites globally. This is a huge ask within 3 years and to minimise the risks to the project we are engaging with a variety of centers that will allow us to test the technology and map accurately the blueprint for how new centers can be brought on board. If we did get the funding we would be delighted to engage with many more centers and hospital networks in LMICs to start initially collecting data for future inclusion in the overall solution. As we build the tool and infrastructure we expect sites to join us proactively and start storing their data and we will happily engage.
In response to What makes your solution innovative and unique?
Potential issues this solution faces.
Scalability, data privacy, and data ownership.
Data privacy: The aggregation of data would mean that the solution would work at the national and regional levels with health systems across the globe, assuming they would comply in sharing health data of their populations.
Data ownership: Sharing to who? If the solution is incorporated as a private entity, I do not see how governments would work with the solution. It would need to be an independent body with credibility antecedents.
Scalability: How to centralize the data is my question. Health systems across the globe have their own structure and include various levels of technology. Epidemiologic risks are greater in LMICs, where countries have limited technology resources and no or limited AI frameworks. To reach them and incentivize them to use the solution can be a lengthy and harduous process.
Thank you for this question, I will answer each below.
Privacy: We are quite insulated from significant privacy concerns as we only need completely anonymised CBC data. We do not need to have any patient information or other EHR data for the algorithm to work.
Ownership: The solution would be held within a global academic consortium between several major centers. This should mitigate some of these concerns. We also have experience establishing these consortia and navigating the issues with doing this.
Scalability: Our proposed solution would be to stream the data to cloud based storage solutions (AWS, Google, etc) directly from centers. The amount of data to be streamed in each packet is extremely small and it would be possible to set up the infrastructure to do this in remote areas if there is any mobile or internet connection. In extreme cases this could be streamed daily if live streaming is not possible. Once the data is streamed all processing is performed in the cloud environment with findings pinged back to local or national public health bodies to flag an event is occurring.
Also, in our work, we are deploying the algorithms in a format that can be computed on just a laptop allowing local computations if cloud access is impossible. Ultimately, deploying inside the machines would be the ambition to remove the need for data to move.
In response to Describe the core technology and/or underlying data that powers your solution.
One question comes to mind: anomalies result in errors when the algorithm encounters R-CBC profiles not seen before but can the algorithm identify the anomaly and pool it with similar profiles which would show with certainty peaks or error from the same blood profiles?
Thanks for the question. Yes, you are absolutely right. In our experiments we have a two-stage process, (1) initially flagging that clusters of people are arriving at GP surgeries / blood donation sites / hospitals that are out of distribution, (2) comparison of feature distribution in these anomalous samples to those 'seen' before, this allows us to determine whether there are similar features in the new CBC results. This allows us to identify biological understanding of the effects of the new pathogen on the blood, over and above that possible with other surveillance mechanisms (like testing sewage).
to Top