Do better predictions result in better decisions?

Karin Frick, 26 May 2014

A health revolution is coming. This much is clear. Through smart devices and smart analytics we will soon have personalised health predictions. What is less clear is what we might do with this information.

A global blackout takes place on 6 October 2010. As a result, everyone sees their future and learns where they will be in six months' time, what they will be doing and with whom they will be doing it. This is the premise of the US TV series FlashForward, which was broadcast from 2009 to 2010. The series was not very successful - perhaps the story was too complicated. But it did examine a question that we will have to deal with increasingly in future, namely: How would we live if we knew a lot more about our future? What would we decide to do, for example, if we could predict our future state of health? And can we change the future by taking a different decision?

The story is based on a novel by the Canadian science fiction writer Robert J. Sawyer, and although it describes a situation that is still science fiction, we are moving closer to the point where this scenario becomes a reality.

FlashForward: Health

Thanks to advances in biotechnology and information technology, we now have more data on our personal health than ever before. This data comes from genetic analyses, which can be carried out increasingly quickly and cheaply, and from the increasing use of mobile devices, which is generating huge quantities of data about people's behaviour.  

While the findings from genetic analyses are already being used for the diagnosis and treatment of a variety of diseases, the potential of smartphone data for health, for example, is only just being recognised. Yet already Google or my mobile device may already know more about my personal health risks than my doctor or insurance company. By analysing searches, Google can predict the course of flu epidemics, for example, or a critical interaction between drugs before any public agency can1.  In a recent article for the New York Times, Eric Horvitz, co-director of Microsoft Research in Redmond, Washington, wrote about software that can predict the risks of post-natal depression with uncanny precision by analysing the tweets of mothers of new-born babies on Twitter and measuring how many times they use words such as "I" and "me"2.  Smartphones will deliver a lot more valuable information on health in the future. We will soon be wearing the next generation of mobile devices such as Google Glass and Apple iWatch. They will record everything we do: who we communicate with; how much exercise we take; what we eat; how we feel; and how we sleep. Functions for measuring parameters such as our pulse, blood pressure and blood sugar levels will soon become part of the standard repertoire of mobile communication devices. The analysis of the communication and exercise patterns of very large numbers of people will allow us to predict individual health risks increasingly well. The more data we have available to us, the better the results will be. Many companies are currently developing health apps to be used specifically for these kinds of predictions.  

For example, Ginger IO is a smartphone app that can predict two days in advance whether the user will suffer depression. The app analyses communication and exercise patterns obtained from smartphone data. Ginger IO focuses primarily on people with diabetes, who have a high risk of depression.

From cure to prevention

Such data is defining a new medical paradigm. Currently patients do not go to the doctor until they have a health problem or something hurts. The doctor examines the patient, makes a diagnosis and prescribes treatment. However, if we know our health risks earlier as a result of progressive improvements in predictions, action can be taken much earlier to treat us or even prevent the need for treatment, ideally long before we become ill. This shifts the focus of treatment from diagnosis to prediction. Prevention will become central and benefit from much better targeting. Instead of just distributing general information about how to live healthily, efforts will be made to target specific risks. The advantage of digital health monitoring is that it enables us to assess an individual's status in the context of external factors influencing that individual, thus taking into account both the individual's physical attributes and current situation. Initially, the focus will be on short-term predictions: people who are chronically ill will be monitored in order to predict possible complications and treat them at an early stage. In future, treatment and prediction will increasingly merge (to become theranostics or personalised medicine). The focus of treatment will gradually shift from pills to apps that continuously monitor an individual's state of health. For example, sleeping pills will be replaced by sleep apps that monitor sleeping rhythms and coach people to ensure they get a good night's sleep. As the Internet develops, homes and clothing will increasingly be equipped with sensors that seamlessly record everything we do, provide real-time feedback and nudge us towards healthier behaviour.

Assuming the predictions of these tools become increasingly more accurate and reliable, and more and more people use them, will they take better decisions, be ill and have accidents less often, and remain longer in their relationships with their partners? What will be the effect of negative predictions? What will people do if these systems predict that they have a significantly higher risk of cancer or Alzheimer's disease?

Patient heal thyself?

The premise of those offering predictive analytics is that the more we know about potential risks, the better we can avoid them. We make predictions because we assume/hope that better information will lead people to take more rational decisions, that improved insight will lead them to avoid doing things that damage their health. But is that true? The results of behavioural research show that people are generally guided by their emotions rather than reason when they take decisions. While positive predictions are likely to enforce current behaviour patterns, literature suggests negative predictions may lead to a number of different responses:

  • Change behaviour. If you know the risk, you can adapt your behaviour and possibly take further preventive action.
  • Overreact/panic. The fear of a potential illness can also lead to overreactions. The people affected may go to the doctor more often, undergo more health checks or be paralysed with fear.
  • Underestimate. Most people are naturally optimistic and believe they will be the lucky exception who does not contract the disease – and that they therefore don't have to do anything to prevent it.
  • Ignore. An unfavourable prediction or prognosis is often completely ignored. People cast doubt on the method, assume there are measurement errors or make the outcome more acceptable to themselves by saying, for example: "Why should I worry myself today about what will happen in 30 years' time. Everyone has to die at some point, don't they?"

Even if the prognosis is relatively simple and clear, such as with smoking or excessive alcohol consumption, the results of which have been predictable for many decades, it doesn't necessarily lead to a change of behaviour. Whether better prognoses result in changes in behaviour is likely to depend on factors such as the following:

The time horizon. The further we look into the future, the greater are the uncertainties, and the further away something is, the less important it is to us. How important is it to a 25-year-old if her life expectancy is reduced by seven years? With predictions about the future, we devalue future events, and that is a significant factor in how we respond to them. Warnings about future diseases are taken less seriously the further away these prospective diseases are felt to be.

Relevance and value. How important is the predicted event to me? Is it important to me that I am still slim and fit in 20 years' time, and what is it worth to me? Am I prepared to do without something now or make an additional effort – save money, take more exercise or keep to a diet, for example – in order to gain a future benefit? Perhaps I will have an accident before I reach retirement age and will get more out of life if I live it up a bit. As the saying goes: "Eat, drink and be merry, for tomorrow we may die."

How probability is experienced. How probable do I think it is that an event will really happen and affect me? Our understanding of statistics and how we assess probabilities are crucial here. When risks are communicated, it is often done by means of percentages. A significant problem associated with expressing probabilities as percentages is that many people struggle to relate those percentages to everyday life.

Emotions, hope and fear. Will life be better or worse in future? Should I fear what is to come, or can I look forward to it? The ability to imagine things has a significant effect on how probability is judged. When events unleash strong emotions, it is very difficult for us to accord them a lower probability. Media analyses have shown that deaths as a result of murders, accidents and natural disasters are reported disproportionately often; whereas deaths as a result of illness are not reported often enough given the probability of their occurrence. Long-term dangers such as climate change do not have this effect, due to the fact that they are constantly there. As a result, we fear improbable events such as plane crashes more than more probable, less spectacular dangers such as car crashes or environmental pollution.

Sense of control. Is the predicted disease curable? Can I do anything to prevent it? For example, there are things we can do to influence whether we develop adult-onset diabetes. Diabetes is often the result of unhealthy eating habits and a lack of exercise. Even if you have a genetic predisposition to diabetes, you can do something to prevent the occurrence of the disease by eating well and taking exercise. It is different with diseases over which we have no influence. Control is also a question of who has the data and who knows my health risks. How will other people and organisations change their decisions about me if they learn that I have a higher risk of suffering from certain serious diseases? Will they support me or exclude me?

If prognoses are to be successful, joined-up thinking and collaboration across different fields is required. Unless the data is networked, predictive medicine will not become a reality. Whether and under what conditions increased self-monitoring will lead to people adopting healthier lifestyles or becoming hypochondriacs is unclear at this stage. What is clear is that monitoring is never neutral. If you know you are being observed, you take different decisions and behave differently from people who feel they are not being observed. We are interested in predictions because we would like to have more control over our lives and be exposed to fewer risks. Whether and how intensively we use predictive and self-monitoring tools in future will thus depend on whether we feel they give us more control over our lives; or whether we fear being controlled and manipulated by a few organisations with too much power.



1; Web-scale pharmacovigilance: listening to signals from the crowd, in: Journal of the American Medical Informatics Association.
2 Eduardo Porter: Tech Leaps, Job Losses and Rising Inequality, NYT, April 15, 2014.

Download full article (PDF 159 KB)


Karin Frick

Head Think Tank, and Member of the Executive Board, GDI Gottlieber Duttweiler Institute

Karin Frick is Head of Think Tank and a Member of the Executive Board at the GDI Gottlieb Duttweiler Institute. As an economist, she has been researching and analysing trends and countertrends in business, society and consumption for over 20 years.

Since graduating from the University of St. Gallen (HSG), Karin Frick has held various positions in which she worked on future-related topics, innovation and change in people and markets. She was editor-in-chief of the well-known quarterly publication GDI IMPULS and managing director of the Swiss Society for Futures Studies (SwissFuture). She analysed developments in the consumer goods and service sectors on behalf of noted companies.

Related articles

Data Analytics in life insurance: lessons from predictive underwriting

William Trump, 26 May 2014

All major companies are thinking about what they can do with 'big data.' Swiss Re has been running data analytics projects with its life insurance clients for the last twelve years. Looking back on his journey, William Trump shares five lessons learned in dealing with data and building predictive models.

Nudging health behaviours through smartphones

Dominic King, Mohammad Mobasheri, 26 May 2014

Millions of people die in both the developed and developing worlds from preventable diseases such as cardiovascular disease and diabetes. By adopting healthier lifestyles and making the best use of the latest evidence-based healthcare, people can reduce their risk of premature morbidity and mortality. Governments and healthcare providers need more effective strategies to encourage healthier choices and increasing interest is being directed at the potential role of new mobile communication technologies as a mode of intervention delivery to help people improve their health.