Over the past few years, the use of artificial intelligence in healthcare and medicine has generated much excitement in both the public and private sector. Just last week, it was announced that AI could diagnose breast cancer from mammograms with more accuracy than a human doctor.
As well as being beneficial in the diagnosis, and potentially the prediction, of individual conditions, big data, and the AI used to analyse it, also has the potential to address public health concerns for the population as a whole.
According to the World Health Organisation (WHO), data analytics could play a potentially important role in “personal care, clinical care and public health, and related research” and has already proved effective in “building accurate models of disease progression and providing personalised medicine in clinical practice”. The organisation adds: “It has also facilitated the evaluation of the impact of health policies and improved the efficiency of clinical trials.”
One organisation investigating how big data can be harnessed to help achieve some of its public health goals is Public Health England, part of the Department of Health and Social Care. The agency is investigating how insights from a combination of consumer-generated data and traditional healthcare data can help improve individual and societal wellbeing.
“We have a volume of data that we’ve never experienced before”
Speaking at Big Data LDN in November 2019, Professor Peter Bradley, director for Health Intelligence at Public Health England, shed light on how the organisation is using big data to inform policy decisions, but also better engage with the public when it comes to bringing about those changes.
According to the European Public Health Alliance, AI has the potential to “improve screening, diagnoses and treatments across many medical disciplines and in many disease areas”, but all of this relies on data. The increasing presence of AI in public health has been driven by access to a greater volume and variety of data, combined with greater computing power. Bradley explains how this differs from the past:
“The data that we’re able to access and use is changing all the time so now we have a volume of data that we’ve never experienced before. Terabytes of data. The data is changing in its timing as it’s going from data where we had a big time lag, maybe an annual survey or a monthly return on hospital episode statistics, to data in real time.
“The data itself is now available in many forms. We have data on the problems that people have, the phenotype, what they’re exposed to, we have this enlarged geographies, and smaller geographies, and we have behavioural data from wearable devices.”
The integration of data from a variety of sources, combining health data and non-health data, meant that it is possible to form a more complete picture of public health. Bradley explains that by analysing data about the population, it is possible to identify trends that can become the focus of future policy:
“One example I want to give is one where we looked at a very broad range of indicators all concerning health and we used two supervised and unsupervised machine learning to look at a pattern of health in England to try and see if there were patterns that we hadn’t really understood before. We found that there was and we’ve called it the London effect…what we found was that London has a different pattern.
“Nearly everywhere else in the country poverty is the thing that drives or deprivation, drives poorer health, but in London that relationship isn’t as clear. If you’re in the poorest parts of London your health is probably slightly worse, if anything, than the rest of the country. When we come to look at the more affluent areas even in the middle of society we find that Londoners’ health is far, far better. This is really important to us because we have the start of an idea of how we might be able to improve health in other regions if we can investigate further.”
However, data does not only need to come from conventional sources, with social media having potential applications in identifying potential public health issues and informing the best way to educate the public about particular issues.
Bradley explains how text analysis has been used to identify the best way to combat misinformation surrounding vaccinations:
“What we’re able to do is look at the Twitter feed. In this case we looked at every Twitter feed that had [vaccine related tweets] and we can begin to understand the sentiment behind these feeds, the types of people who are showing concern. That allows us to target messaging so that the appropriate scientific advice is given to groups to counteract this myth that has persisted in in the society around vaccinations.”
Personalised health advice has also been deployed to help people quit smoking. Demographic targeting has been used to help Public Health England deliver ads tailored to the individual, making it more likely that they will successfully quit.
Looking to the future
While artificial intelligence has proved effective in identifying patterns or trends in data, it is now moving into the realm of prediction. Identifying future public health events, such as predicting the volume of people that will use A&E services is important when it comes to planning and allocating resources:
“So far we’ve based a lot of our analysis on what’s already happened. And what we’re trying to do is move into the position where we are trying to predict more of future health problems so we can take action before they actually happen. But sometimes even now the retrospective analysis is very meaningful.
“In the predictive elements we’re actually trying to predict the future so that we can use our resources more appropriately and make sure that services are geared up to deal with the demand that will come. This example is from NHS and it’s to do with a number of attendances you would expect in an accident and emergency departments. We used [programming language] R to create a model that predicts the number of attendances and what we found is that this model is very accurate. We’ve obviously tested it with the real data and we now feel we’ll be able to predict the number of accidents in emergency by the hour for several years in advance and this is really important obviously for NHS planners and making sure that the demand meets the supply.”
Bradley believes that, when deployed correctly, big data can be harnessed to give patients a far more personal approach to healthcare:
“We’re now moving towards a possibility of having a much more personalised approach where we can improve health by using the data that’s coming through, big data potentially, and making sure that we are using all those possibilities that we have in terms of using data from smart phones and citizen generated data in all its forms.”
“We don’t want to be prejudicing people who we are most trying to help”
Looking to the future, Public Health England is now working towards a more personal approach in how data is used to drive engagement and improve health on an individual level, better harness data from wearable devices and smart devices, and combine this better with other forms of data, such as GP’s records. However, when handling data as sensitive as health records, a number of considerations around ethics and privacy arise.
Bradley gives the example of the NHS health check, which is given to those between 40 and 74 to check for signs of some diseases. He envisions that in the future this could be conducted through the NHS app by individuals entering their data, and would then be offered strategies to improve their health, with activities tracked through a wearable device.
Bradley explains the importance of ensuring that any deployment of technology keeps the agency’s core goal of keeping the population safe and improving general health:
“We are going to have to think very closely about what the public really wants to share in terms of the data and what do they want to use it for. We already work very closely with the National Data Guardian and other groups that are concerned about information governance and this is a really big area for us.
“We can’t currently access a lot of our data because it’s not deemed appropriate so the need for that debate is really crucial…artificial intelligence and related activities can introduce bias so the way that we train our data is very important. We don’t want to be prejudicing people who we are most trying to help.”
Another issue arises when it comes to the public’s trust of Big Tech when it comes to health data. According to ZDNet, the NHS database contains the primary care records of 55 million people, making it an understandably attractive resource for any company, but generates controversy when data is shared incorrectly, such as an incident in 2017 when data from the Royal Free Hospital was shared with Google’s DeepMind.
However, Bradley believes that, when done correctly, collaboration with other parties is important:
“The health data environment is actually quite complex and it does involve us using many data types and I think as we move forward collaboration with the academic and the commercial sector is going to be crucial. I can’t imagine we’re going to be able to do this on our own. We do need to find ways that we are really keeping up with the latest approaches.”
Lucy Ingham 2:12 PM