It's science, but not as we know it. Thomas Granatir looks at the prediction methods that could change the face of public health.
Millie is 87. She is overweight. She has type-2 diabetes and congestive heart failure. In recent years, she has developed severe vascular degeneration and painfully debilitating peripheral neuropathy. She walks with great difficulty.
Millie is well educated. Her health literacy is first-rate. Yet she fails to control her weight, eats irresponsibly and takes her medication irregularly. This renders her condition unstable and dangerous. If it were not for the care she receives from husband Willie, she would be in a nursing home or dead.
Willie is 90. He spends 30 minutes a day doing floor exercises he started doing when he turned 40. He goes to the gym three times a week in his relentless fight against sarcopenia. He weighs what he did 50 years ago. Willie does the shopping and cooking and takes care of Millie. In the past year, his only serious medical event was a cracked tooth.
Millie's condition is the result of behaviours that have persisted for 60 years. But there is a subtext. the healthcare system that has done such a good job of keeping her alive is helpless to deal with the underlying behavioural issues that have conditioned her health.
The cost of treating acute episodes for 'Millies' will threaten the viability of health systems across the globe. The growing incidence of behaviour-induced disease - obesity, cardiovascular disease, cancer, and diabetes - challenges our health systems to rethink how they deliver care, and our public health systems to rethink how we manage health.
Humana is a large US health insurer with responsibility for about 11 million lives. About six years ago, we began to approach this problem by finding and influencing people careering towards poor health.
To do this, we had to figure out who is going to be sick, how sick, and how soon. We had to stop thinking about people as clinical objects and try to understand the complex social, psychological, and cultural contexts of the decisions they make that render them more like Millie and less like Willie.
The science we drew on was more computational and anthropological than clinical.
Our primary, secondary, hospital, laboratory and pharmacy data is organised into a person-centered data warehouse. Once we organise the data around the individual, we can see the care paths of each over time. These can be compared with evidence-based practices to identify people who may be receiving too little care, too much, or the wrong sort. By modelling at the individual level, we can describe health trends aggregated to a GP practice, hospital, locality, or region.
Finding high users in this data is easy. But how do we find those 'at risk' who may be high users tomorrow?
Most risk modelling looks at clinical events, translating them with the help of clinical algorithms into a risk score that represents future medical costs and use. But what if we assume that patients are not inert objects? What if we assume biological, psychological, social, cultural and environmental variables - that cannot be reduced to a set of clinical algorithms?
What if we assume these variables are interdependent; that a small change today can mean a large change tomorrow? These assumptions seem obvious, but they are the opposite of those made in conventional health analytics.
The third way
Computational science exploits advances in IT and maths so scientists can work on problems previously thought too difficult or too large. It allows scientists to build models that predict what might happen in the lab, which help them understand what they then see in real experiments.
But it is also used to perform experiments too expensive or dangerous to conduct in the lab, such as finding out how a new drug might behave in the body.
About five years ago, Humana began using signal processing techniques from the engineering sciences to analyse health data. These are different from most clinical modelling in that they treat clinical events as signals of a change in the individual's health state.
With these techniques, we can find Millie earlier and target interventions to mitigate her risk.
To help people, we have to know how they live and understand their medical problems in context. In patient-generated clinical encounters, context is discovered through a few questions asked by the clinician. In the new world of analytics, it is understood through the analysis of data - as much as we can get, and much more diverse than in the past. Our ability to understand people's context means we can characterise their circumstances, experience, behaviours, and how these change over time.
At Humana, we link encounter data to other signals, such as online logs, calls to a demand management line, and personal health risk assessments. These can tell us if a person is worried or anxious about a clinical matter.
With the new methods, we can incorporate new kinds of data and ask new questions - a kind of 'lifestyle analytics' - to help lifestyle change. In one pilot, for example, we are collecting physical activity and grocery data to see how these increase our ability to predict health risk.
Biosensors will be next, capturing heart rate, blood pressure, stress levels, blood glucose, and more. Activity monitors that record daily physical activity are available.
What it will mean
These changes open up the possibility of an entirely new approach to public health.
The next generation of public health professionals will understand populations through a deeper understanding of individuals. They will graduate from a piecemeal, ad-hoc, patient-initiated care system to systematic prevention and health monitoring.
This approach will make it possible to move clinical management out of the clinic and into patients' life stream, where health creation takes place.
Thomas Granatir is policy and research director at Humana. Europe.