I was fortunate to have been invited to attend the recent ‘Data & Analytics in Healthcare’ conference in Australia which took place at the end of March. The event was positioned as a ‘deep-dive into the data transformation taking place in healthcare across Australia – and aimed to explore efforts to transform the use of data, analytics, and AI in order to deliver great efficiencies and better, more accurate, patient care’. In my blog this month, I will explore whether this was achieved and my key takeaways from the event.
What was interesting about the three opening scene-setting presentations is that they all highlighted the need to operationalise insights, to make those insights actionable, and have insights being used in the actual process workflow. I think healthcare, more than any other industry, suffers from the challenge of having amazingly useful insights being generated, published, and talked about, but very few organisations actually manage to operationalise those insights. This is especially the case when it comes to making those insights practical for people ‘on the floor’, like administrators, clinicians, specialists, and educators, to improve business or patient outcomes.
Education driver
A related theme that came through strongly was the need to educate business users and decision-makers on the meaning of and how to make good use of the information and insights. We must keep in mind that making sense of data insights is not something healthcare practitioners are trained in. Their focus is on improving patient outcomes.
We would therefore need to educate the practitioners on the insights generated. This should include even straightforward things like how to access and run a dashboard, what a specific indicator means, and so on. Presenting the information in a useable format also makes it easier to understand and act on it. I have posted several times about data storytelling and visualisation – those aspects can be used productively to enable better utilisation of insights in healthcare.
Making sense of data
Several sessions highlighted the challenges that must be overcome. This provided some insights into why it is so difficult to operationalise analytical insights. Most healthcare organisations reported digitally immature operational systems. Many of these organisations stated that their data was siloed across numerous systems, with much data duplication and many manual processes in between.
A significant number of systems were developed by external providers who were mostly internationally-based. This meant the service providers had different operational contexts within their solutions that are extremely difficult and costly to customise.
Healthcare data might not be voluminous, but it is often wide, varying, and complex. Some healthcare data is still recorded manually on pen and paper and then transcribed into the respective systems. There is also a lot of unstructured and textual data in healthcare – significantly more than in most other industries.
Governance structures
Various governance frameworks were discussed as were ethics and aspects around the privacy of patients’ sensitive data. The overarching approach was to use techniques such as masking, de-identification, and other statistical approaches. This would make the data useful for research, case studies, and collaboration to improve other patients’ outcomes but without ever allowing the identification or disclosure of an individual’s personal and private data.
There was even a presentation on the governance around insights models that are made available in a shareable forum. I found applying governance to a calculation, a formula, or a complex model an interesting concept. Of course, the model has to be validated and trusted, and must include error and exception handling. But I still think if you apply a model to an incorrect or skewed or inappropriate dataset, you will get misleading results which may be detrimental to the outcomes.
Introducing AI
There were also interesting discussions around machine learning and the use of artificial intelligence like ChatGPT. If I had to summarise these into one statement then it would be that when we use such advanced technology to make decisions about how or with what exactly to treat a patient, we are stepping into a no-go zone where technology is misused for purposes it should never be intended for.
However, there are many examples where these technologies can be used to vastly improve the process around triage, diagnosis, follow-up monitoring, and more. As an example, I would not like ChatGPT to recommend a treatment for whatever I am suffering from. I would, however, be happy if the clinician were to use the technology to find, summarise, and prioritise 12 research papers on the topic to enable them to make a more informed diagnosis. There are some very interesting applications in post-treatment and age-care monitoring and alerting as well.
In closing
After a hiatus of more than two years, it was good to attend a face-to-face event, catch up with old colleagues, network with new connections and broaden the horizons with what is happening and what is possible in a very comfortable and collaborative environment.
I must also give kudos to Corinium for a well-run event and good use of technology – it was totally paper-less, with some interesting partners and vendor displays. There was also wholesome healthy food and snacks and a friendly good quality coffee barista in the house to keep the energy levels up!
