Health data access will transform health outcomes for all… the clock is ticking

While propagating reasonable concerns about access, the recent debate about the opt out timing for My Health Record, has however exposed a terrible lack of informed input into what could be the greatest positive change in health outcomes for Australians… ever.

Artificial intelligence and machine learning will transform healthcare diagnosis globally

The case in Australia 🇦🇺

The simple story is that for decades, maybe hundreds of years, Patient A (let’s call her Mary) has:

  • Fronted up to her local doctor with her symptoms and asked for a diagnosis and treatment
  • Over time doctors have increasingly sent Mary off for a battery of blood tests, an MRI, and CT scans to try to diagnose her disease conditions that may not have been immediately obvious
  • In some (most) cases, this has resulted in a correct diagnosis and treatment selection
  • Then months, maybe years later, Mary experiences new symptoms and trots off to see the doctor again
  • If she sees the same doctor as before, then maybe the doctor will refer to her history to see if the current condition is similar to previously diagnosed conditions
  • However, if Mary goes to a different doctor then she will not necessarily have access to her “old” health data, and the doctor is asked to make a diagnosis with no clear sense of what happened previously to Mary

Such a process can lead to terrible consequences and the only way to avoid this is to ensure that when Mary goes to a doctor (any doctor)… that doctor has access to all of Mary’s health history.

By having this access, the doctor has the best chance of a correct diagnosis. However, this is ONLY possible if Mary did not opt out of the MHR data repository.

Sure, we need to require that:

  1. The Government puts the same level of security on our health data as banks do with our financial data (I think they are)
  2. If someone does hack the data then the data can’t be used to bias a decision against someone.

The case in the US 🇺🇸

This “misuse” issue was addressed in the US by making modifications to their discrimination legislation with regard to Genome sequencing —

For example: If you had your genome sequenced and you happen to carry a gene that is involved in say Breast Cancer, then a health insurer can NOT use this data to bias your health premiums. Nor can an employer enact any biases against you on the basis of this data.

We need to make sure this regime is in place in Australia with regard to any health data.

Better access to health data = better health outcomes

Moving to the BIGGER issue… how do we use the massive repository of health data to help individuals have much better health outcomes?

Going back to dear Mary, even the best doctors can only deal with a certain level of data input and multiple variables… humans simply run out of puff.

We need quite strong signalling to make decisions.

In health this is best expressed as:

  • A blood test (or biopsy) that shows you have, or are very likely to have, a disease
  • Or an MRI that shows an appropriately sized lump that can then be biopsied to test for cancer (however, the lump needs to be big enough to show up for humans to see!)

…bring on software.

Machine learning (ML) algorithms can take large (low noise) data sets and “learn” to diagnose a disease. These ML algorithms can learn to diagnose disease conditions more accurately and often much earlier than humans can.

A few (of many) high profile examples of how machine learning is changing disease diagnoses

  • MetaOptima using machine learning to diagnose melanoma
  • Google using eye scan data as a diagnostic tool for cardiovascular disease

The list of current projects where companies are using anonymised health data sets to train algorithms is long, and getting longer every day.

Most countries have worked out that if you can develop the world’s first and best diagnostic tool for X disease then the benefits are enormous.

Using the cardiovascular disease example:

  1. Millions of people will be diagnosed earlier, have less interventions and not die
  2. Health systems will (due to early intervention) save a ton of money and health systems will be able to offer universal high quality diagnosis to everyone
  3. You can sell these tools globally and therefore create economic opportunity for the company and country of origin

Currently there is a lot of discussion about how to “tap” the NHS data sets (in an anonymised manner) to allow the creation of a vast array of health diagnostic tools.

The opportunity we have in Australia

Interestingly, there are really only a couple of countries that have very large, “all of nation”, health data sets — which provide the data on all health interventions. These countries have a leg up on the rest of the world in terms of the race to develop diagnostics tools.

🇬🇧 UK: Even though Deepmind (now part of Google) was founded in the UK, the reason they have such a large research presence in the UK is due to the NHS data set.

🇦🇺 Australia: We are one of a couple of other countries that have access to these massive data sets. In our case, we have Medicare data on most of the 24m residents of Australia, and with the PBS data set, we know all the prescription drugs prescribed to Australians. In health data terms this is gold, and gives Australia a massive leg up in the race.

We have the opportunity to take some of the most prevalent disease conditions (think diabetes, cardiovascular disease, and a whole bunch of cancers) and their patient data sets and create the world’s best diagnostics tools.

We can then deploy these tools across Australia to ensure every single Australian has access to quality diagnosis — as if every Aussie was sitting in a room with the best doctor in the world.

Then we can sell these tools overseas.

However, this is not an opportunity that will exist forever, because when you first create a diagnosis tool you:

  • Need large (low noise data sets)
  • Then get a bunch of very smart Data Scientists (and some folk with domain expertise in the disease condition you are narrowing in on) and off you go developing the learning algorithm
  • Over time the algorithm becomes very good. Soon it becomes near perfect
  • Then one day you don’t really need any more data to train the algorithm. Vast data sets are not really of that much value any more as the algorithm has learned all it needs to.

This means that our massive advantage — our large health data sets — is not an advantage that we will have forever.

My guess is if we do not start to develop diagnosis tools that are globally best in class for each disease condition… then, in a few years we will wake up and all our doctors will be using overseas developed diagnosis tools. And Australia will have missed the greatest opportunity for national wealth creation in a generation (or more likely a few generations).

“So what now?”

We need to understand that the world is not waiting.

Every government around the world understands the opportunity (and the threats) that AI/ML will bring, and we are in a race to see who will be the Google of melanoma diagnosis, cardiovascular disease diagnosis, and diabetes diagnosis… you get the drift.

What do we need to do:

  • Get over ourselves… do not opt out of MHR
  • Understand that if you want to get the benefit of future improvements in health, then the “cost” to you is your data being on a secure government data repository
  • Ensure we have appropriate data misuse laws in place
  • Ensure we have robust access regimes for the Medicare, PBS and large pathology data sets. Anonymised of course!
  • Promote greater collaboration around research groups working on disease conditions.
  • Make sure the grey nomads running the various medical colleges (i.e. college of surgeons etc.) get on the AI/ML train.. Yes. I am sorry to say... computers can do some things better than you. Get over it. Use these tools to make better treatment decisions.

And now.

As individuals we have the ability to help enable all Australians to lead healthier lives… We will do this by enabling our health data to be used to create new diagnosis tools and interventions that lead to less pain and suffering.

Organisations have a responsibility to firstly advocate for what is the future of healthcare and the implications around data use, while also making sure we actually do no harm to any individual, family or group.

Let’s get over ourselves and move towards a healthier future for all.


I’m a partner at Airtree Ventures, one of the largest venture capital funds in Australia, and actively looking to invest in HealthTech companies that are changing health outcomes for patients and saving money and saving lives. I’ve previously written about this here and here.

If you have any comments on my post — daniel@airtree.vc or dpetre. Would love to discuss further!