Photo credit: Frank Lindecke

By Dr Trishan Panch

The saga involving Cambridge Analytica and Facebook is alarming not because it is a breach in the conventional sense: where a nefarious hacker exploits a security vulnerability to steal sensitive information but rather because it is a breach of trust – a symbolic betrayal of an implicit contract around privacy and independence of thought.

Though the Facebook terms of service seemed to permit data sharing of this nature, many users were not aware that information as specific as religious affiliation and sexual preference were being shared with third party app developers and then sold into a shrouded network of marketers working nefariously to advance their own ends through targeted messaging on social media. The revolution no longer needs to be televised it seems.

Many of us find it alarming that, based on personal information, we have been drip fed messages to alter our points of view at the same time as our friends and neighbors are in turn targeted with the self-same techniques. All of this to advance a particular political candidate or outcome by exploiting our need to be connected to each other.

I believe that it is the objective as much as the method that is the issue here. The techniques used are powerful and could be an incredible force for good if deployed for a more genuinely productive objective. Like making people live longer.
I am an optimist in all matters related to computation. I believe humans and computers (in a general sense) can and will work together to make the world a better place. I appreciate the near-term risks regarding privacy and protection of independent thought as well as the long range risks of machine doomsday.

A lot of the same underlying techniques used in political campaigns and consumer marketing are part of the long run solution if applied appropriately and with respect to the privacy and unique needs of patients.

My somewhat optimistic view is that even if that transpires, we will have made some transformative breakthroughs on the way: a more complete understanding of the inheritable component of disease and when and why we should intervene as well as a deeper understanding of the mechanisms underpinning environmental factors that modify and determine the expression of our code, efficient food production and logistics of all varieties to name but a few.

Already, of course, computers and process engineering are, slowly, making healthcare better. But much more can and should be done, and a lot of the same underlying techniques used in political campaigns and consumer marketing are part of the long run solution if applied appropriately and with respect to the privacy and unique needs of patients.

Furthermore, the techniques Cambridge Analytica used to optimize messages are incredibly manual and outmoded and can be considerably improved with more granular data and machine learning methods.

I am particularly interested in the vast majority of healthcare that occurs outside of the hospital or clinic for patients with existing chronic conditions who, I believe, create their own “health”. As clinicians, though we like to put ourselves at the center of the universe, for these patients we are merely the suppliers of some of the raw materials and some of the knowledge necessary for that creation.

In truth, we perform our function in a crude manner that would be totally unacceptable in any other manufacturing process. In addition, patients are, of course, people and have preferences broader than just their clinical needs.

Each person in a health system is a Turing machine of sorts — each of us going through a series of health states where the transformations are the actions of the health system.

Whilst individual clinicians are supremely skilled at understanding and adapting to these consumer preferences, in the health system individuality tends to be ignored in favor of more of a one size fits all approach where every patient is treated as an average patient.

I believe, irrespective of their background or health condition, people should know what they need to do and look out for to make the most of their health (assuming health is a measurable quantity which it may or may not be — but for the purposes of this argument I am assuming it is). I believe in some instances it is possible to turn the existing ossified one size fits all model on its head using techniques of data informed population segmentation, targeted healthcare interventions using digital products and data infrastructure with machine learning to create closed loop feedback systems.

These are descendants of the techniques used in political advertising but when applied with skill and respect can deliver tangible clinical and financial results for providers, payors and better, more connected care for patients.
The computational element of this is actually more tractable than people believe and in fact the construct of a Turing machine is an interesting way to think about the health of an individual and health systems.

I was so inspired by the power of Alan Turing’s formulation of a Universal Turing Machine – a general model of computation not linked to the materials of computation, that I wanted to call our first daughter after him but I could not find a female derivative of his name that worked (Alanis was torpedoed by my wife’s antipathy to Morisette).

Each person in a health system is a Turing machine of sorts — each of us going through a series of health states where the transformations are the actions of the health system. In this way a hospital (or arguably any unit of healthcare delivery) is as much in the business of information processing, of computation in a general sense as it is in the business of care.

Now of course I am not saying the caring elements don’t matter (I spent 10 years trying to demonstrate that they do) nor that medical knowledge (the truth inside the aforementioned transformations) is anything like perfect but for the ease of this argument it is more convenient to simplify things a little bit.

The future of preventive medicine will not be one of humans or computers but humans and computers – each operating where they have differential advantage and delivering a whole that is greater than the sum of its parts.

What is left are two core computation tasks: diagnosis and management. Diagnosis, in the relatively data-poor community setting, is principally constrained by the lack of the kind of structured data available in the ICU or in radiology, a sensing problem first and foremost, really in the basic science realm.

In most cases we need better sensors to bring some objective comparability to what is otherwise subject to the vagaries of how people use language to describe things that are happening to them and to others.
With plentiful structured data the diagnosis problem becomes a simpler issue of classification where the profound leaps in machine learning are already being demonstrated.

I think the management problem is much more tractable in the short term. This is largely an issue of sensing the health state of a patient or a group of patients and determining the most appropriate intervention to get the patient to the next best health state on the path to an ideal health state — a health goal.

Digital Health interventions for interactive chronic disease management create both the necessary data to do this and the adaptable mechanism of intervention. I see the techniques of data science for segmentation and targeted personalized, machine optimized clinical interventions as new tools of clinical quality improvement.

Crucially this is also better for patients as this combination of techniques delivers personalized, adaptive, evidence based preventive medicine to everyone across a whole country whilst also freeing up clinician time to help patients with more complex needs.

The future of preventive medicine will not be one of humans or computers but humans and computers – each operating where they have differential advantage and delivering a whole that is greater than the sum of its parts.

Computers are incredibly good at disseminating, collecting and processing vast amounts of similar information and now, with machine learning, making some sense of the world. Human clinician time saved through automation could be spent on the activities that people are really good at and computers are really bad at — making other people feel cared for.

Whilst the techniques of segmentation, targeting and iterative optimization used in consumer marketing can be used for nefarious ends the potential for benefit is also considerable.

In order to realize the benefits in a way that works for patients and for health systems, it is important that a new breed of clinicians are involved in the development of this powerful new set of digital products and infrastructure with more respect and consideration for the needs of people than has been displayed in other industries. Ultimately, progress will not be achieved by replacing care but by amplifying it.

 

This article is dedicated to my uncle, Sithamparanathan Sivagnanam who died a year ago. He was a fool who dreamed and on such things is all human progress based.

 

Trishan Panch preventive medicine artificial intelligenceDr Trishan Panch is a primary care physician and Co-Founder and Chief Medical Officer of Wellframe Inc. He works at the intersection of medicine, computer science and entrepreneurship, developing products and services for digital transformation of payor and provider organisations in the US and Europe. He is an advisor to MIT Critical Data, MIT lecturer in Health Sciences and Technology and is the inaugural recipient of Harvard’s Public Health Innovator of the Year award.
He is the holder of US Patent No. 9,805,163: Panch et al. “Apparatus and Method for Improving Compliance with a Therapeutic Regimen” which details a generalizable model for using mobile technology and artificial intelligence to dynamically organize health care around an individual patients needs while simultaneously achieving system level optimization in a way that is applicable to all clinical verticals.
He currently serves on the board of Wellframe and the Innovation Advisory Board of Boston Children’s Hospital. His work founding Wellframe was the feature of Harvard Business School MBA Program case 816-062 by Professor Bob Higgins. He was previously an NHS GP Principal in London and trained in medicine at Imperial College.