Blood Counts!

3.6 billion Complete Blood Counts are performed globally each year and can be used to detect infectious disease outbreaks.

Carola-Bibiane Schönlieb, Professor of Applied Mathematics and Head of the University of Cambridge Image Analysis Group, Department of Applied Mathematics and Theoretical Physics, LMS Whitehead Prize recipient.

Early detection of infectious disease outbreaks is paramount for mounting effective public health measures. Current approaches used for infectious disease surveillance rely on prior-knowledge of the pathogen and, as illustrated by the current pandemic, they are inadequate for early detection of novel threats.

There is an urgent need for simple, affordable and scalable methods that can be applied in High, Middle, and Low Income Countries (HIC, LMIC) which can detect infectious disease outbreaks in real-time and are pathogen-agnostic. 

Our solution to this problem is to use the data from the 3.6 Billion Complete Blood Count (CBC) tests performed globally each year for infectious disease outbreak detection. After reporting of high-level summary results, Rich CBC (R-CBC) laser measurement data is usually discarded. Since Jan-2019 we have rescued 2.8M R-CBC measurements performed on the Greater North-London and Cambridgeshire population (6M). We have developed Machine Learning models which can detect the local SARS-CoV-2 outbreak in a pathogen-agnostic manner using this data. We postulate that our solution will also work for other pathogens, including novel zoonoses, without prior-knowledge of DNA/RNA sequence. With a CBC test costing £1.10, scaling surveillance in HM-ICs can be readily achieved using our method.

Our solution will provide health protection organisations with a sensitive, low cost method for early detection of infectious disease outbreaks in the populations they safeguard. Drawing on the experience of our team members, who have developed infectious disease screening technologies and innovative population health measures (Leung K., et al., Lancet, 2020) and cutting-edge AI-ML solutions (Tso A.K., et al., Brain, 2021), we outline requirements below:

1. Public health organisations need globally accessible technologies for monitoring infectious disease outbreaks, including ones caused by novel pathogens.

2. Detection must occur as rapidly as possible to deliver effective infection control.

3. If the technology is to be used as an early warning system, it cannot be reactive; population monitoring and event detection should rely as little as possible on prior-knowledge (e.g. pathogen DNA/RNA sequence).  

4. Resources are always limited. The technology must be cheap, ideally already funded, simple to use, scalable and require limited logistical support to deploy.

5. Data analysis and storage requirements must be minimal, especially if the technology is to be used in LMICs, which often lack national data networking infrastructure.

We seek to provide an affordable and globally accessible method for infectious disease surveillance and public health protection. With this in mind, our solution makes use of data that has already been paid for and returns additional value in the form of improved health protection. To ensure global benefit, the technology we are proposing is simple to use, requires no prior-knowledge of the pathogen for analysis, is cheap, does not require freezers, and the compute requirements for data analysis are small. 

Lastly, much more information is contained within the R-CBC measurement data then can be discussed in this application. Aggregation and analysis of this data at scale will bring other tangible benefits to society, with improved methods for disease detection, including heart attacks, stroke, autoimmune diseases and blood cell cancers. As an example, our team has used R-CBC data in the past to define the genetic landscape of blood cell traits (Astle et al., Cell 2016; Vuckovic et al., Cell,  2020).

The 141M infections and 3.4M deaths that have occurred during the SARS-CoV-2 pandemic to date demonstrate that even HICs have found it challenging to implement effective public health protection measures at population scale. For example, it took Public Health England months to make PCR testing for SARS-CoV-2 available at national scale after facing challenges in method validation, equipment availability, and supply constraints on laboratory plastics. 

Our solution provides a method for early detection of novel and known pathogen outbreaks, with testing coming at no extra cost and only limited resources requirements for data capture, streaming, and analysis. With 10M CBC tests performed each day, we believe that our solution will provide public health organisations with an additional early warning system for local detection infectious disease outbreaks that will inform quicker and more specific deployment of infection control measures. 

Lastly, CBC measurements are a cornerstone of medical practice. Increasing the availability of CBC instruments, particularly in LICs where infectious outbreaks frequently originate, not only provides an additional layer of public health protection, but will also improve the quality of healthcare in resource-limited settings.

The CBC testing component of our solution is already available in all HIC’s and in a large proportion of LMIC’s. We must now focus on scaling data collection and analysis (Fig.5).

Year-1: 

i) Complete pilot study using 2.8M previously captured R-CBC measurements. ii) Expand analysis to 10M R-CBC measurements performed by NHS laboratories connected through UKRI Microsoft Azure cloud computing platform iii) Develop consortium data governance framework to HIPAA standard or equivalent. 

Year-2:

i) Capture R-CBC measurements for a population of approximately 35M individuals through our international haematology network of 97 Hospitals and Laboratories (Turro et al., Nature 2020) ii) Trial at-scale streaming and analysis of R-CBC data to secure analysis environment iii) Formalise international partnerships between data infrastructure providers, instrument manufacturers, and public health agencies to prepare for global scaling.

Year-3:

i) Engage with regulatory authorities and leading CBC instrument manufacturers Sysmex, Beckman-Coulter and Siemens to prepare for global roll-out to obtain approval for embedding ML models in CBC instrument control software.  

ii) Capture R-CBC measurements from international partners in India, Hong Kong and Singapore expanding R-CBC measurement capture to a population of 305M. 

iii) Scale rescue of R-CBC data to LMICs by working with other Trinity Challenge partners. 

Year 1:

1. Sensitivity - are 0.5%/1%/5%/... of CBC tests required to be affected for detection?

2. Specificity - can we identify SARS-CoV-2 outbreaks amongst other infectious events (e.g. Flu season)? 

3. Speed - how quickly can we detect increases in localised case loads? 

4. Scale - governance in place for aggregation and centralised analysis of R-CBC data.

5. Impact - number of R-CBC rescued, number of R-CBC data aggregated centrally, successful engagement of LMICs.   

Year 2:

1. Scale - number of CBC tests/month for which R-CBC data are automatically streamed to the centralised analysis platform; ratio between H, M and L -ICs. 

2. Impact - sensitivity for detection of seasonal increases in infectious disease case loads; R-CBC data volumes analysed and made available to Public Health Authorities.

3. Accessibility - Challenges we have faced in implementing our solution in countries not named in the application; Have methods been developed to overcome any identified barriers? 

Year 3: 

1. Scale - Coverage of 305M population in HIC, MIC and LIC.

2. Impact - Has CE-marking / FDA regulatory approval been obtained?

3. Accessibility - has application been incorporated by CBC instrument manufacturers for automated alert systems?

Data collection poses a challenge. CBC instruments are clinically regulated instruments and the environments of attached ‘analysis computers’ are usually ‘locked’. As a result, many are not networked and use outdated operating systems making on-board analysis not feasible. We managed to overcome these challenges in three large-scale CBC-testing laboratories using Microsoft's “SyncToy” in combination with manual data recovery via hard disk. 

To globally scale our solution, resources are required to help CBC test centers develop automated methods for transfer of their data to analysis environments. Palantir is a global leader in this area. Their expertise will be instrumental in developing data capture solutions suitable for the ‘real world’. 

Having access to R-CBC measurements of the healthy population is a major challenge, which must be overcome, because of these data being required as a baseline. This challenge can be overcome by using our existing Consortium of 8 National Blood Services (Gleadall et al., Blood Advances, 2020), covering a 330M population, and performing 6.8M CBC tests/year on their donors.  Legal and ethics agreement for data transfer and aggregation are in place and the analysis platform is operational.  

Universities of: Amsterdam, Cambridge, Leuven, University College London, Hong Kong, Singapore; Apollo Hospitals, India; Hospitals affiliated with above Universities, Barts Health Hospitals London, National Blood Services of Australia, Canada, Finland, Ireland, Netherlands, New Zealand, South Africa; Palantir Technologies

To deliver this project it is essential to straddle across academia, healthcare providers and the private sector. The possible award from the Trinity Challenge would provide a framework for this. Through the Trinity Challenge we aim to leverage support from other Data Analytics companies to achieve global scale deployment. Furthermore the Trinity Challenge founding members can provide a knowledge framework for infectious diseases (including, for example, the Bill & Melinda Gates Foundation and the IHME) would give a mechanism for expanding the solution equitably across the globe.

The applicants would need support in connecting the laboratories which provide CBC testing in different countries. Such labs may be hospital-based or are highly centralised. Interestingly Sysmex has over 75% of its instruments connected to a central reporting database for calibration. Could this existing network be exploited to stream the R-CBC data from interested parties.  

If accepted to the challenge, we recognise the importance of bring the R-CBC analysis results to other parties with similar anonymised data, from which outbreaks could be inferred (e.g. Google, Facebook). We could foresee a role for the Bill and Melinda Gates foundation to support LICs to increase their capacity for CBC tests. In particular Africa, disproportionately to its population size, performs a very low number of CBC tests. Increasing availability of venesection and CBC instrument capacity could bring tangible benefits for clinical medicine and public health protection, including for malaria and AIDS detection and prevention.  

The Trinity Challenge could facilitate connecting the three main manufacturers of CBC instruments to the challenge of developing and commissioning an affordable ‘early detection test’.