Cough In A Box (CIAB) - UK Health Security Agency, Data Futures Division

The UK Health Security Agency is an executive agency of the Department for Health and Social Care (DHSC) (hereinafter referred to as “UKHSA”) located at 39 Victoria Street, London, SW1H 0EU. UKHSA co-ordinates across the UK, building strong collaborations with public health agencies for Scotland, Wales, and Northern Ireland, and operates internationally for the UK to help understand, prevent and respond to global threats to health. DHSC is the data controller in respect of the data collected by, and further processed within, the constituent parts of the organisation.

The Data Analytics and Surveillance Group operates as a specialist data analytical, assessment and research provision within the UKHSA. The Data Futures division is a function that has been carried into UKHSA from the Joint Biosecurity Centre for the purpose of carrying out research and innovation projects such as 'Cough-in-a-Box' and Environmental Monitoring for Health Protection.

The ‘Cough-in-a-Box’ (CIAB) project aims to develop a process for the identification of changes to vocal patterns as a symptom of Covid-19. This has the potential to be a fast and easily administered early test which would pave the way for targeted mass testing. It also has the potential, in due course, to be applied to the early detection of other diseases.

It builds on early-stage research from Massachusetts Institute of Technology (MIT)/Harvard (https://ieeexplore.ieee.org/document/9208795), which reported the ability of an algorithm to accurately classify asymptomatic COVID-19 positive patients using audio recordings of forced coughs from a small participant cohort. The use of this algorithm, for instance distributed at scale in a smartphone app, has potential as a rapid and affordable COVID-19 screening tool.

The project aims to collect an independent research database of audio recordings linked to COVID-19 test results, and to use this data to build and evaluate algorithms that may detect COVID-19 from audio recordings.

This research study follows a No10 commission to assess the potential to screen for COVID-19 using vocal biomarkers, following publication of initial proof-of-concept research by academic groups in MIT (https://ieeexplore.ieee.org/document/9208795) and Cambridge (https://arxiv.org/abs/2006.05919).

The study has three different groups of participants who will be invited to take part in the research.

One pool consists of everyone who has had a positive or negative COVID-19 test in England. This pool will be contacted via the Agile Lighthouse teams (hosted by Sitel), using contact details from the Pillar 2 Antigen testing dataset provided by NHSD under Data Sharing Agreement NIC-406871. When the patient receives their test result, they are asked if they want to participate in further research and if they say yes then they will also receive information about the CIAB research and will be directed to the study’s website (ciab2021.co.uk). When they reach the website, they will be provided with a privacy notice (available at https://ciab2021.uk/run/survey3.cfm?CFID=cc308556-1853-474a-9c31-7a0debf5127a&CFTOKEN=0&idx=505d040f0b). If they accept that they have read and understood the privacy notice they will then be provided with some more information about the study and asked if they agree to the recordings and supplementary information they are about to provide being processed for the research purposes that have been described.

Participants also fall into this pool if they are not contacted at all, and hear about the study from other sources, e.g. word of mouth etc.

The voice recordings and other information provided by the patient via the webform will then be linked to the patients' test result in the Pillar 2 Antigen testing dataset provided by NHSD under NIC-406871. This is the only pool of participants whose CIAB data is linked back to the NHSD data. The data is linked using the participant's test bar code.

Another pool consists of patients participating in the REal-time Assessment of Community Transmission 1 (REACT 1) study. NHS Digital have provided Ipsos MORI with identifying data from the Patient Demographic Service (NIC-393650), the contact details are used to contact a sample of people to ask them to register to take part in the REACT 1 study. Participants are also asked if they agree to be contacted to take part in further research. If they have agreed to be contacted for further research the patient will receive an email from Ipsos MORI containing information about the CIAB research and will be directed to the study’s website (https://ciab2021.uk/REACT). When they reach the website, they will be provided with a privacy notice (available at https://ciab2021.co.uk/run/survey3.cfm?CFID=cc308556-1853-474a-9c31-7a0debf5127a&CFTOKEN=0&idx=505d040e08). If they accept that they have read and understood the privacy notice they will then be provided with some more information about the study and asked if they agree to the recordings and supplementary information they are about to provide being processed for the research purposes that have been described.

The voice recordings and other information provided by the patient via the webform will then be linked to the patients test results in the REACT 1 dataset held by DHSC and Imperial College London. The CIAB data is not linked back to NHS Digital data.

The final pool consists of patients participating in the Human Challenge (HC) study

https://www.ox.ac.uk/news/2021-04-19-human-challenge-trial-launches-study-immune-response-covid-19.

If they have agreed to be contacted for further research the patient will receive information about the CIAB research and will be directed to the study’s website. When they reach the website, they will be provided with a privacy notice. If they accept that they have read and understood the privacy notice, they will then be provided with some more information about the study and asked if they agree to the recordings and supplementary information they are about to provide being processed for the research purposes that have been described.

The voice recordings and other information provided by the patient via the webform will then be linked to the patients test results in the HC study dataset. The CIAB data is not being linked to NHS D data.

In all cases, the participants read and confirm understanding of the privacy notice and additional information and consent to taking part in the study. The participants provide voice recordings which will be used to develop and assess an algorithm for the purpose of screening for COVID-19. The aim of the study is to assess accuracies of algorithms in detecting COVID-19 from voice audio recordings.

The CIAB study inclusion criteria are:

- Residing in the UK

- Over the age of 18 years

- Speak the English language

- Taken a COVID-19 test within the last 72 hours (regardless of test result)

- Access to a mobile device with internet connection.

Although this is a government funded national study, there are some pragmatic reasons which have meant that these exclusion criteria must be implemented at this early stage of research. Exclusion criteria based on age are a consequence of capacity to provide informed consent to participate in the study and ability to engage with the webform. The inclusion criterion requiring the ability to speak English is due to limited study team capacity to develop a webform in multiple languages at this stage. Finally, the data collection approach currently limits inclusion criteria to only those volunteers with access to the appropriate digital technology with an internet connection.

The results of this study will not be generalisable to those age groups or spoken languages excluded from this study. Additionally, the study team recognise that the requirement for technology literacy and access will lead to an unknown degree of sample bias, and extrapolation of the study findings to the general population may be limited by this.

This is an early stage research study to explore the potential application of machine learning approaches to classify COVID-19 disease based on audio data. Significant validation work, separate to this research study, would need to be carried out before this technology could be implemented for use in the community. There would need to be a particular emphasis on ensuring that this work validated findings from this study among groups underrepresented in the study cohort who may be targeted as part of any future use case for this technology. It would be essential to ensure that this validation work demonstrates that this technology had been developed to work effectively for all target groups across society, particularly so whilst recognising that this is a government funded initiative.

The legal basis for processing personal information for this project is that it is necessary for the performance of a task carried out in the public interest in that the outcomes will help towards preventing the spread and incidence of COVID-19. Personal information will, therefore, be processed according to Article 6(1)(e) GDPR, and all special category personal information will be processed according to Article 9(2)(j) GDPR. All such processing will be proportionate to the aim pursued, respect the essence of the right to data protection and provide for suitable and specific measures to safeguard the fundamental rights and the interests of the data subject.

The part of the cohort that consists of those with a positive test result are contacted under explicit consent, which they can choose to give when they are contacted as part of the Contact Tracing. Sitel Ltd also carry out part of this Contact Tracing through their call centres and as such already have the positive contact details (via the Contact Tracing and Advice Service (CTAS) dataset from PHE).

Sitel Ltd will be processing the contact details for those in the cohort with a negative test result from the Pillar 2 Antigen testing dataset (provided by NHSD) under COPI as it is not possible to gain explicit consent from those with a negative result due to them not going through CTAS.

Being able to process contact details for those who have recently engaged with Contact Tracing Services under COPI, rather than solely under the explicit consent obtained when participants receive a positive result from a COVID-19 test allows the study team to contact those who receive a negative COVID-19 test result in the same way as is already being conducted for those who receive a positive test result. This approach reduces the risk of bias in the cohort.

The Department of Health and Social Care are determining the means and purpose of the processing of the data and are the data controllers. The data processors are the Department of Health and Social Care and Sitel UK Limited (Agile Lighthouse teams).

This CIAB research study is needing to collect data from volunteering participants in order to assess the accuracy of algorithms to detect COVID-19 from voice recordings. Once a strong enough algorithm has been found there would be a separate piece of work to develop the method for distributing this new way of testing, e.g. an app, but that is not in the scope of this research study.

Patients, in the pools described in section 5a. ‘Objective for processing’, will be approached to volunteer to participate in the data collection study, where they would be linked to a web-form they would complete on their smartphone or other device. Following a privacy notice and acceptance form, the web-form will ask the participant to answer several questions about their symptoms, existing health conditions, smoker status, and provide their COVID-19 test barcode number, after which the form will prompt participants to record several short audio recordings including a forced cough, breathing sounds and a scripted sentence.

The part of the cohort that consists of those participants with a positive test result are contacted under explicit consent, which they can choose to give when they are contacted as part of the Contact Tracing. Sitel Ltd also carry out part of this Contact Tracing through their call centres and as such already have the positive contact details (via the Contact Tracing and Advice Service (CTAS) dataset from PHE).

Sitel Ltd will be processing the contact details for those in the cohort with a negative test result from the Pillar 2 Antigen testing dataset (provided by NHSD) under COPI as it is not possible to gain explicit consent from those with a negative result due to them not going through CTAS.

Sitel Ltd comply with the relevant provisions of UK GDPR and Regulation 7 of COPI. The DHSC as the data controllers satisfy the requirement in Regulation 7(2) COPI. DHSC have a contract in place with Sitel Ltd who are carrying out the processing under the instruction of DHSC and makes it clear that it is DHSC who has overall control of what happens to the personal data. In the contract it is clear that confidential data must be treated in such a way to meet the Regulation 7(2) requirement (they owe a duty of confidentiality, as per Regulation 7(2)), staff within Sitel Ltd who are handling the data are aware of this fact and are subject to contractual obligations of confidentiality. These provisions satisfy the requirement in Regulation 7(2).

Data to be collected from research Study Participants via the web form;

• Audio samples (.wav format, 20 seconds max duration)

• Test Barcode ID

• Symptoms

• Smoker Status

• Respiratory Health Conditions

• First Language

• Height

• Weight

• Whether they are wearing a mask at the time of recording

The web-form has been developed and managed by a third party procured by DHSC (currently Studio24), who will temporarily store participant information and recordings in an Azure server before they are transferred to DHSC systems, where they will be linked to COVID-19 test results using the test barcode number and analysis will take place. This third party will not store or process any NHS Digital Data. They will only be processing data derived from the web form.

Substantive employees of DHSC are able to work remotely on DHSC issued devices only, the data must not be downloaded to a local device and the data can only be accessed remotely within the territory of use.

The CIAB submissions will then be linked using the Test Barcode ID to the test results data relevant to the pool in which they were invited through. Only the Agile Lighthouse participants CIAB data will be linked to NHS D data.

The test results for the patients coming through the first pool (Agile Lighthouse) described in section 5a are in the Pillar 2 Antigen Testing dataset held by NHS Digital. A subset of this dataset for the CIAB participants will be shared from the DHSC secure Data and Analytics Environment (EDGE) to a secure DHSC Project Environment.

There will be 2 subsets of this dataset used for the project.

The first subset will be linked to the participants CIAB data using the NHS Test barcode and is set out below;

- ttcepseudo - This is a pseudonymised identifier created within EDGE using the field NHS Number. Enables duplicate/repeat submissions made by individuals to be linked together within our data set, to reduce risk of the model overfitting the data.

- Ethnicity - Enables us to identify any bias in our data gathering, and validate if the model works effectively across different ethnic groups.

- Local Authority

- Date of Onset of Symptoms

- Gender

- Age (from DoB - transformed in EDGE before transfer)

- TestReason - Enables us to identify any bias in our data gathering.

- VaccinationStatus - Enables us to validate if model works effectively across groups with different vaccination status.

- VaccinationPeriod - Enables us to validate if model works effectively across groups with different vaccination status.

- ProcessingLabCode - Enables us to better interpret Ct values (due to different calibrations/machinery) and filter results by lab.

- Symptomatic Indicator

- Test Type

- Test Result

- Sample Created Date

- Specimen ID

- Specimen Processed Date

- CH1 Result

- CH1 Target

- CH1 Value

- CH2 Result

- CH2 Target

- CH2 Value

- CH3 Result

- CH3 Target

- CH3 Value

- CH4 Result

- CH4 Target

- CH4 Value

This will be filtered to only records of patients who have submitted their data via the CIAB (identified pseudonymously through NHS Test barcode) before leaving the DHSC Secure Data and Analytics Environment (EDGE).

365 days following the end of the project the Test Barcode ID will be replaced with a random identifier to anonymise the data set.

Before sharing any of the data (e.g. with academics for further research, who will have honorary contracts in place) the Test Barcode ID will be replaced with a random identifier to anonymise the data set.

Reaching out to patients in this first pool to invite them to volunteer will be done via the Agile Lighthouse teams, hosted by Sitel. These teams are the teams that already contact patients as part of Contact Tracing operations. Sitel are required to use a subset of contact details from Pillar 2 Antigen testing data that will be provided to them by UKHSA following the permission given under this agreement.

The privacy notice for NHS testing states:

"If you test positive or negative, you may also be contacted by DHSC to see if you wish to contribute to the research effort of COVID-19. If you are interested in doing this, you need to follow the link in the text message."

"Your information may also be used for different purposes that are not directly related to your health and care. Wherever possible, this will be done using information that does not identify you (anonymous data). These include:

- research into COVID-19 (including potentially being invited to a research project)"

(This privacy notice can be found here, https://www.gov.uk/government/publications/coronavirus-covid-19-testing-privacy-information/testing-for-coronavirus-privacy-information--2)

The subset of the Pillar 2 dataset to be processed by Sitel in order for their lighthouse teams to invite patients to the study is:

- Name - required to engage individuals by call

- Phone Number - required to engage individuals by call/text

- Date/time the test occurred - required to ensure we're only contacting individuals within the eligible timeframe for submitting vocal recordings.

- Specimen ID - required to provide to individuals so that they can complete the CIAB form (as they may not have kept a record of their test barcode).

The filters that will be applied in EDGE will be as follows:

- Test Result - Only negative results.

- Test Type - Only PCR tests.

- Pillar - Only Pillar 2 tests.

- Phone number - Only records with phone numbers.

- Kit registered date - Only kits registered over the previous day (i.e. added within the last 24 hours).

- DoB/Age - Only individuals 18 and over.

This subset of data will not be combined in any way with any other data held by the project.

In addition to this, the project will also be using other sources of testing data related to the two other pools described in section 5a, REACT and HC participants. This data will be linked to the submissions of recordings and supplementary information provided by patients from those pools. These other sources of testing data will not be combined in any way with the testing data in the Pillar 2 Antigen testing dataset being shared under this agreement.

The research study is supported by partners based at the Alan Turing Institute (ATI). They will be supporting with the analysis work for the project. Members of this research group are onboarded to DHSC systems following pre-employment checks and agreement to the onboarding declaration and will have honorary contracts in place.

Patient and Public Involvement

The CIAB project has currently carried out no patient and public involvement (PPI). However, there are plans being developed by the project team to begin involving patients and public rather than simply just informing them. As the project moves closer to product development there will be an ever-increasing focus on PPI.

To develop and assess an algorithm for the purpose of screening for COVID-19. When a strong enough algorithm is found this will be used to develop an app or similar that will be made available to the general public in order for them to test whether they have COVID-19 or not. However, the development of the app is not in scope for this research project and would be a separate piece of work following further validation work.

The results of this research will be published in peer reviewed journals and presented at relevant academic conferences. For example, The Lancet Digital Health has published related works, and study results would reach an appropriate audience spanning the life and computing science fields through this journal.

Results may also be disseminated through other approaches such as reports or briefings on behalf of the DHSC and collaborators or via DHSC and collaborators' communications channels.

Results of the study will also be reported to stakeholders (such as other Government departments, health research organisations and Academic Institutes) through briefings to the data science campus at No10 Downing Street, as well as presentations to study stakeholder groups and innovation teams within the UKHSA.

As this is a Government sponsored study there may be media interest in the results of the study. Any responses to requests for comment from the media regarding published results will highlight whether or not the findings have been peer reviewed and make every effort to ensure appropriate representation of the study in 3rd party communications to the public. Study sponsors will not influence the study team’s interpretation of the results.

Study outputs will report aggregated data with small number suppression applied in accordance with ONS Guidance for statistical outputs.

This is a research study to develop and refine models to classify COVID-19 disease based on clinically validated audio recordings of cough/voice/breath. As such, it is not a clinical study of a new screening tool. Outputs from this work will likely be limited to research study findings (such as sensitivity and specificity of the model), and the code base for the data pre-processing and analysis pipeline of this work to be made publicly available where possible in line with a commitment to transparent and replicable working. Researchers based at the Alan Turing Institute (ATI) involved in the analysis work for this project will draft academic articles and publications in collaboration with JBC study team members. Members of this ATI research group have been onboarded to DHSC systems for this work and DHSC remains the data controller for data collected. These members will not be allowed access to the NHSD data until they have honorary contract with DHSC. All outputs (aggregated with small numbers suppressed) will be made openly accessible where possible.

Target dates for outputs are dependent on rate of prospective data collection (audio data) for the study and study progress. Final results from the study are expected in March 2022, dependent on sufficient data collection.

Sharing of pseudonymised health data for this purpose will facilitate the research of technology which has potential for a rapid, minimally invasive, low-cost screening tool as social distancing regulations are released following the greatest peaks to date of the pandemic in the UK. Currently mass testing in the UK is supported by semi-invasive swabs as part of relatively expensive tests (estimated £3 – 10 / LFD test). Machine learning approaches to classify COVID-19 disease have potential to reduce the need for more invasive swab-based screening initiatives and make these programmes more affordable. There is further potential in this approach when it is recognised that only around 20% of symptomatic cases engage with swab-testing services. A more rapid, less invasive tool could begin to increase engagement of this missing cohort from screening initiatives.

This study represents early-stage research. If the results of this study indicate the potential of this technology is evidenced, the project will be taken forward for further research and validation before any future consideration regarding implementation. This validation will emphasise work with groups underrepresented in the current study and will be implemented by engagement with policy stakeholders associated with 10 Downing Street and the innovation teams within UKHSA’s Data Futures division. Outputs from this research will be made open access where possible for transparency and replication by others.

If this study demonstrates the usefulness of this machine learning approach to screening for COVID-19, it could have potential as a rapid, minimally invasive, low-cost screening tool as social distancing regulations are released following the greatest peaks to date of the pandemic in the UK. Currently mass testing in the UK is supported by semi-invasive swabs as part of relatively expensive tests (estimated £3 – 10 / LFD test). Machine learning approaches to classify COVID-19 disease have potential to reduce the need for more invasive swab-based screening initiatives and make these programmes more affordable. There is further potential in this approach when it is recognised that only around 20% of symptomatic cases engage with swab-testing services. Working in collaboration with behavioural scientists to contribute to a multi-faceted approach to addressing barriers to engaging with testing and tracing services, a more rapid, less invasive tool could contribute towards increasing engagement of this missing cohort from screening initiatives.

This research is funded through a DHSC awarded grant, not carried out in support of a PhD or post-graduate research study.

PPI

Online versions of the participant information sheet is on the .gov web page. Where participants may find out about the study electronically, they are directed to this. Participants recruited in-person at test sites are provided with a paper version of this. The study materials, design and recruitment approaches were approved by NHS REC. The Privacy notice makes clear the data linkage between NHS Digital datasets and primary data collection through the Cough-in-a-box study.

Not stated in the register.