A blog by Dr Susan Thomas and Professor Matthew Thompson from Google Health.
Why do we need novel types of data to help improve diagnosis?
For most health conditions, earlier identification and diagnosis results in better treatment options and outcomes for individuals. However, for many of the most serious conditions we face (for example, cancers, or serious mental health disorders), doctors and other healthcare professionals have struggled to identify reliable ‘early warning’ signs. The absence of these types of signals makes it hard – both for doctors and patients – to know what to look out for, and therefore when and, as importantly, when not to worry.
Most research into early symptom detection has typically relied on information from people’s contacts with the healthcare system, usually through their GP or hospital records; but this can mean, in many cases, that by the time individuals seek care, subtle symptoms may have been present for some time thus limiting their treatment options. And even if people do seek help when their disease is in an early stage, symptoms are often non-specific and difficult to interpret or may be mistaken for other less serious illnesses.
Identifying more reliable and earlier signals of serious conditions would give patients and doctors greater predictive abilities and help them to better differentiate serious from non-serious diseases.
Reaching ‘back in time’- can online searches help?
What if we could ‘look back in time’ to help us solve this problem? If medical researchers could better understand what an individual was worried about and experiencing in the weeks or months before their diagnosis, would that help us understand how to catch serious illnesses sooner? Could looking at what people were searching for online help to unlock insights into the ‘pre-diagnosis’ phase of illness development?
Research using GP and hospital records has definitely provided hugely valuable evidence about early symptoms of many diseases, but this data is only available after someone has been to their GP or hospital visit.
For acute, rapidly developing illnesses (acute infections for example) asking people (or their loved ones) to try to remember what they were feeling, after they have been diagnosed, can work due to the short timescales. But for slower onset diseases, such as many cancers or long-term conditions, people’s memories and recall of their feelings or events from months or even years prior are inevitably less reliable.
What if someone’s online search activity could provide us with insights?
In the UK, almost 50 million health-related searches are made using Google per year. Globally there are 100s of millions of health-related searches every day. And, of course, people are doing these searches in real-time, looking for answers to their concerns in the moment. It’s also possible that, even if people aren’t noticing and searching about changes to their health, their behaviour is changing. Maybe they are searching more at night because they are having difficulty sleeping or maybe they are spending more (or less) time online. Maybe an individual’s search history could actually be really useful for researchers. This realisation has led medical researchers to start to explore whether individuals’ online search activity could help provide those subtle, almost unnoticeable signals that point to the beginning of a serious illness.
Our recent review found 23 studies have been published so far that have done exactly this. These studies suggest that online search activity among people later diagnosed with a variety of conditions ranging from pancreatic cancer and stroke to mood disorders, was different to people who did not have one of these conditions.
One of these studies was published by researchers at Imperial College London, who used online search activity to identify signals of women with gynaecological malignancies. They found that women with malignant (e.g. ovarian cancer) and benign conditions had different search patterns, up to two months prior to a GP referral.
Pause for a moment, and think about what this could mean. Ovarian cancer is one of the most devastating cancers women get. It’s desperately hard to detect early – and yet there are signals of this cancer visible in women’s internet searches months before diagnosis?
Where next for this novel data?
These studies show that individuals’ online activity is a new and potentially valuable form of data for healthcare research. However, this type of data is appropriately private and cannot be used by health researchers without explicit consent. For many years now, Google has made it easy for users of its products, such as Search and YouTube, to download their own data using a free product called Takeout, giving individuals control over how they access and share their data. This means that health researchers can ask individuals if they are happy to share their data – for example, their Google Search history, as part of a medical research study in exactly the same way they might provide consent to use their medical record data.
While the initial studies published using online search data have been highly encouraging, many questions remain. We don’t yet know the added value of online searches as part of diagnostic or prognostic research, how generalisable published findings are, and most importantly, how this research could be implemented to help patients (and their health care providers) in real-world clinical practice. However, given the types of health challenges patients and clinicians struggle to diagnose, it seems worth pursuing to both understand the utility of these early signals and to establish the robust clinical evidence needed.
Want to learn more?
Using this type of data is new to most health researchers. To help answer questions and provide resources that researchers can use, Smart Data Research UK and Google Health hosted a webinar. During the event, we shared some of the science behind using internet search data. The panel included Dr Srjdan Saso, who has used online search activity to identify signals of women with gynaecological malignancies.
You can now watch the recording of this great discussion.