Apps and algorithms to help predict illness: Many of these applications fall into the “lifestyle and well-being” category of products, but they nonetheless indicate a trend which will change medicine. With big data, medical treatment will become more personalized, more preventative, more proactive.
Health apps for Smartphone are booming. Around 100,000 such apps already exist, meant to help with weight loss and to mitigate depression, to calculate fertility cycles, or to train the user in mindfulness. At the same time, new sensors are constantly being developed: fitness wristbands and smart watches count steps, monitor sleep and measure heart rates. Cameras, rings, patches and implantable sensors measure skin conductance, perspiration and blood values. Google, Apple, Microsoft, Samsung: in recent years all the big IT players have been bringing to market health applications for home use.
This is because lifestyle, fitness and health data applications have developed into a huge market in recent years. They form the missing piece of a puzzle that can perhaps make good on the promises of “personalized medicine” made a decade ago. At that time, the human genome had just been decoded. Using the genetic code, it was said that it would be finally possible to discover treatments for cardiovascular diseases, cancer, or Alzheimer’s. Success, however, has thus far remained elusive. Direct causal relations between genes and illnesses are hard to find, and our genome, so far as we know, works in a much more complex way than we had assumed.
Unbelievable volumes of data
Since then, not only has computer performance drastically improved and the cost of gene sequencing fallen enormously, but there are now unbelievable volumes of digital data available, gleaned from patient records, studies, and, not least, the plethora of health, lifestyle and fitness apps. As people collect more and more data on themselves, and as the number of connections and patterns emerging from this data increase, each individual can more precisely trace their own biological makeup.
“Just as the microscope made things visible which were much too small for the human eye”, wrote American economist Erik Brynjolfsson a few years ago, “the analysis of large volumes of data by means of algorithms makes connections visible which previously were far too big and complex for human understanding.” But lifestyle data, or the personal, health-related data collected by many fitness apps is not easy to relay and aggregate. At least, for the time being, not all of it is. Researchers worldwide are already working on programs that can reveal the complex relationships between body, environment and behaviour and simulate how patients will react to treatments, as well as assist in developing personalized medical interventions.
At the paediatric oncology clinic in Homburg, Norbert Graf is working together with mathematicians, molecular biologists and biological computer scientists to develop a computer model for Wilms’ tumour. This childhood kidney cancer, the professor explains, forces doctors to choose whether to operate immediately or to first treat the tumour with a course of chemotherapy in the hope of shrinking it, so as to render the surgery more straightforward. But not all children respond equally well to chemotherapy.
The program aims to generate a prognosis based on data about the previous development of the tumour, medicines and their active ingredients and the widest possible range of clinical information on the patient. “We want to know how the tumour will respond to prior treatment. Ultimately the system should say: ‘the tumour won’t get any smaller, operate immediately’.” The bigger the volume of data on which the model can draw, and the more frequently its predictions can be measured against outcomes and adjusted accordingly, the more precise its prognoses will become.
Providing the best treatment right from the start
It would be immensely useful for doctors if it were easier to cross-reference data from medical records with personal information—and additionally with genetic test results and studies on the efficacy of different medications—, according to Norbert Graf. Many of his colleagues agree. “That way, we would be able to provide patients with the best treatment right from the start, and reduce the side effects they suffer.” Since 2011, clinics in several European countries have been working to network their databases, and to store information on, amongst other things, illness-related genetic and biological markers in blood and tissue samples. This has resulted in the the Biobanking und Biomolecular Resources Research Infrastructure (BBMRI).
In the USA President Barack Obama provided around 215 million US dollars for the Precision Medicine Initiative, which he inaugurated at the start of this year and which will see the genetic and health data of over one million Americans saved and made available for cross-referencing. This initiative should make it possible to perform tests in order to predict the effects of drugs. Analysis of this database should not only provide hints on how a treatment should be designed to battle an acute illness. The fact that this information is also linked to lifestyle data is “an incredible treasure trove” for medicine, says Norbert Graf, because it can also provide information on the likelihood of relapse.
Graf continues, “Following a successful course of cancer treatment, you always want to avoid a relapse. ‘Is there something special I should eat?’ is a common question, as is ‘Should I do more sport?’ And if I had, for example, information from this kind of health tracker about patients’ sports and nutrition, and if I had long-term information about who had or had not had a relapse—then I would be able to say to someone: ‘if you do this, or if you eat that, you’ll have such-and-such a chance of avoiding a relapse.’ We can’t do that yet.”
A data protection nightmare
Nonetheless, this development is a nightmare from the perspective of data protection. On the one hand, the quality of data recorded by wearable devices and trackers frequently falls far short of medical standards. Studies have repeatedly shown that such devices can often generate false readings. On the other hand, critics fear that the storage of health data cannot be deemed sufficiently secure to guarantee anonymity. One fear is that this could lead to discrimination or disadvantages for those seeking employment, for example, should employers become aware of illnesses or predispositions to certain illnesses. Critics are also worried that in the future it could become obligatory for one to gather data on oneself using various trackers or apps, for the purpose of providing it to doctors or insurers.
Even now, insurers like the German public health insurance AOK or the Swiss Generali Versicherung have started rewarding customers with bonuses and discounts if they can prove they have a healthy lifestyle with data gathered by app. “Currently, it’s all voluntary”, says doctor and e-health expert Tobias Neisecke. “And it’s about rewarding someone who is being proactive about taking care of their health data. But it is probable that this could be turned around. At some point it will become about: ‘what’s my app score?’”
Health insurers insist that there is no disadvantage for members who decline to take part in this health monitoring. Nonetheless, though it remains an open question, bigger business will probably be made with the data itself; it will provide raw material for prognosis models which calculate health risks, not only with a view to creating treatments which are appropriate for target groups, but also for the purpose of developing preventative interventions.
Targeting and speaking early on with at-risk patients
Since 2014, the Carolinas HealthCare System, a network of doctors in the state of North Carolina, has looked at correlations between consumer data and health data in order to identify patients who are at risk for specific illnesses. In Germany, the Elsevier Health Analytics think tank is working on algorithms which can look for patterns in anonymized health insurance data and identify groups of policy holders where there is a given probability that certain illnesses will arise. Doctors will be able to check their patient data against this filter and speak with at-risk patients early on.
The German health insurance provider AOK is also developing a “cardiovascular risk assessor”, according to Kai Kolpatzik from the AOK Federal Association in Berlin. It should predict “how high your risk is of having a stroke or heart attack over the next ten years, on the basis of age and blood pressure, whether you smoke, and your family’s medical history. And what’s exciting is that this can tell you things like: What will happen if I take this medication? What effect would a change in lifestyle have?”
Analysts calculate that if current double-digit annual growth figures persist, the market for personalized medicine will have a global turnover of 90 billion US dollars by 2023. This is money that should belong to the people who provide the data, says Ernst Hafen of ETH Zurich. Together with colleagues, he has initiated the MiData project: a co-operative whose members—patients and health professionals alike—are able to upload genetic and other health-related data onto a server, but decide for themselves what the data can be used for. Companies that use the data must pay for it. The proceeds are to be used to finance research projects which big private firms see as unprofitable.
Apart from the question of who will carry out medical research in the future and who will benefit from it, the predictive analysis of this data is bound to change medicine: instead of diagnosing acute illness, the question is increasingly one of predicting the likelihood of problems occurring down the road. “We are no longer just sick or healthy”, says the medical ethics expert Peter Dabrock, “we are the carriers of given risk profiles. And that’s where it becomes ethically and economically interesting, because that poses a whole new array of questions in terms of the consequences that this has for health insurers. Today, we say: carriers of a given genetic mutation, for example, have a claim for a given treatment, which we pay for. Soon, it could be: We’ll pay for a treatment with 70 percent chance of success. But what about 65 percent? Will we still pay for that?”
Latest posts by Lydia Heller (see all)
- Predictive healthcare: Medicine in the data revolution - 13. July 2017