Behaviour: from intervention to diagnosis

Alessandro Guazzi, Co-Founder at Sentimoto. Alessandro has provided us with an introduction to his presentation on the novel ways of measuring and representing behaviours allowing the description and qualification of complex diseases using behavioural analytics. Particularly, conditions common to ageing which includes rheumatoid arthritis, depression, and frailty.

With 2017’s freshly minted New Year resolutions, comes the annual peak in Google searches for “losing weight” and the doctor warnings against radical “detox” methods. Although at first sight these may seem like trivial vanity blips as people realise just how much they have been drinking and eating, but they point to a much deeper trend.

The effects of lifestyle factors such as smoking, poor diets, and inactivity are progressively making their way into the collective understanding of health, no longer seen as just limited to fixing bones or taking pills. Tackling these behaviours can have as radical an effect on the public’s health as the newest miracle drug, as all are tied to the top killers in industrialised countries. What’s more, making people healthy through behaviour interventions is thought to save a lot of money in the long run. This has led to a general roll-out of preventative schemes that have many similarities to vaccine programmes, but use behaviour - rather than a pill or a jab - as the treatment.

But what about measurement? In the general push towards a data-driven society and evidence-based medicine, it would be odd to continue to rely on patient-based reports and “expert clinical opinion” alone.

Enter the “management” apps.

The computer first, and the smartphone later, apps have helped people manage and change their lifestyles through digital diaries, smoke prevention applications, calorie-counting applications, and now even disease-specific applications such as those for diabetes. Associated hardware companies, such as FitBit and Apple, are also trying to make their way into consumer health, as evidenced by their integration into private insurance wellness programmes. These data-points, whether entered manually or automatically, are gradually making their way into the hands of clinicians.

The ultimate goal is however to use the data not only to measure the progression of a preventative treatment, but to obtain a diagnosis for disorders - quantifying disease symptoms through behavioural data. This is something that has been happening recently in the field of mental health, where behavioural symptoms are some of the best indicators of the severity of disorder. For example, Sentimoto’s CTO Maxim Osipov used physical activity to identify periods of depression and mania in patients affected by bipolar disorder - the first time an objective framework for the analysis of mental health symptoms was developed.

A similar framework may be applied more broadly, and at Sentimoto we are particularly interested in using technology to treat, measure, and diagnose age-related conditions. These are conditions that worsen rapidly with age, such as arthritis, dementia, and cardiovascular disease. The effect of lifestyle on health is particularly important in older people, with many different but interconnected behavioural factors coming into play. A surprising example is that of social isolation, with strong research evidence that loneliness is an independent predictor of mortality in older adults: as bad as smoking 15 cigarettes a day.

Understanding the user’s typical behaviour at any given instant in time - whether they’re physically active, sleeping, or socialising, already means that emergencies such as falls or heart attacks can be spotted early. However, by using a holistic approach to well-being - understood not only as the physical and physiological, but psychological and social as well, we can aim to use consumer technology to both prevent age-related disorders and diagnose them early in a low-cost and effective way.

The analysis of such complex multidimensional data requires using advanced data analysis techniques, capable of dealing with highly complex and (possibly) sparse inputs. What’s more, a range of possible conditions may be responsible for any particular behavioural pattern. In order to ensure the accuracy of the analysis, it has to be validated on large cohorts of patients and controls with different symptoms manifestations - the perfect playground for deep learning at the cross between social care and health care.