By Siu Hing ( Siu Hing ) Lo, on 12 December 2014
Big Data is burgeoning. A quick search using Google Trends shows that worldwide interest took off exponentially since 2011 and is still on the rise. There also seems to be consensus that Big Data has huge potential to improve health services (Accenture Industrial Internet Insights Report for 2015). Of course, the use of large health datasets has a long tradition in epidemiology and other public health disciplines. However, the sheer scale, variety and complexity of Big Data means we increasingly rely on artificial intelligence to manage and analyse data. Simultaneously, there is a trend towards tailoring and personalising health care services, often facilitated by the increasing availability of personal data and more powerful analytical tools.
Big Data certainly has the potential for gaining valuable insights. However, it could also be a double-edged sword, especially in the case of health care tailoring. Yes, Big Data could advance health risk ‘profiling’ and enable more cost-effective ways to tailor health services. But as Khoury and Ioannidis put it, the promise of Big Data also brings the risk of “Big Error”.
Possibly the most obvious caveat lies in the critical interpretation of data (Susan Etlinger at TED talks, 2014). Data does not speak for itself. People need to make the leap from data to insights. One major challenge is how to understand the data. With its ever growing complexity, both the data and the analysis are more likely to be biased in a way that the human interpreter had not foreseen. In other words, the unknown unknown. Another related issue is researcher bias in interpreting the results. What do numbers really tell us about other people’s needs, preferences and perceptions? As complexity increases, fewer people will be able to familiarise themselves sufficiently with the data and analytical methods used to critique study results.
A less obvious source of bias is the type of information sources that are relied on. Of course, Big Data is to be welcomed if it can yield useful insights. However, the promise of Big Data might overshadow the use of smaller scale (e.g. survey, qualitative interview, ethnographic) data and experimental studies. Research funding is finite and popular trends could unduly influence allocation of resources to studies proposing to use Big Data.
The question is whether Big Data can live up to its promise without skilful, complementary use of other methods of inquiry. Despite all the advances in speech recognition and robotic engineering, human beings are generally still much better at understanding other people. Especially in the case of health service tailoring and personalisation, it seems that caution is warranted. After all, if Amazon’s algorithm suggests a book you don’t like, little is at stake and you can ignore the recommendation. But if an algorithm suggests a bad care plan, it would be a less trivial problem, especially if you cannot ignore it. Multidisciplinary collaborations addressing the same public health inquiry would be important for the “Big Picture” as well as the avoidance of “Big Error”. Although multidisciplinary collaboration is not always easy, different viewpoints can facilitate critical and reflective thinking, key ingredients for any meaningful research and truly personalised health care.
To give a personal example, two of my earlier blog posts describe the behavioural insights I derived from studying records of 300,000 people invited for bowel cancer screening in the South of England. Perhaps reflecting wider trends, research psychologists like myself are using ever larger datasets to study people’s health behaviours and identify target groups for behaviour change interventions. However, one question we could not address in these studies was why people behaved the way they did, even if we could predict what they would do. Although these studies did not use Big Data, they illustrate the challenge we face using Big Data or any type of pre-existing data generated for purposes other than research. In order to improve public health, observing the status quo alone is not enough.
Big Data has the potential to yield powerful insights for health service tailoring and personalisation. However, the process of arriving at these insights may pose considerable challenges. Critical thinking and the involvement of researchers who do not typically work with Big Data will be key to its effective use as a tool for health care research.
Accenture (2014), Industrial Internet Insights Report for 2015, Available at http://www.accenture.com/SiteCollectionDocuments/PDF/Accenture-Industrial-Internet-Changing-Competitive-Landscape-Industries.pdf
Khoury, M.J. & J.P.A. Ioannidis (2014), Big data meets public health, Science, 346, 1054-1055.
Lo, S.H., Halloran, S., Snowball, J., Seaman, H., Wardle, J. & C. von Wagner (2014), Colorectal cancer screening uptake over three biennial invitation rounds in the English Bowel Cancer Screening Programme, Gut, Published Online First: 7th May 2014, doi:10.1136/gutjnl-2013-306144.
Lo, S.H., Halloran, S., Snowball, J., Seaman, H., Wardle, J. & C. von Wagner (2014), Predictors of repeat participation in the NHS Bowel Cancer Screening Programme, British Journal of Cancer, Published Online First: 27th November 2014, doi: 10.1038/bjc.2014.569.
Susan Etlinger (2014), What do we do with all this Big Data?, Filmed September 2014 at TED@IBM http://www.ted.com/talks/susan_etlinger_what_do_we_do_with_all_this_big_data/transcript?language=en