By Greg Horne, SAS
We see the most promising applications in healthcare when real time decision making using simple data sets can relieve the burden on the caregiver and drive better outcomes for patients. For example, alarm fatigue in hospitals - both clinically and operationally - is a cause of lost productivity and decreased patient safety. AI technology can be applied to a smart pump to only sound an alarm when there is a critical need for intervention and to initiate an escalation process if there is no response.
Other promising applications include:
- Reducing medication errors by preventing drug contraindications and adverse indications - including the monitoring and correct usage of antibiotics that is driving anti-microbial resistance (set to become a leading cause of death in the next five years);
- Improving mental healthcare and treatment outcomes and applying evidence-based responses; and
- Enhancing decision making to palliative care practices to improve end of life care.
Trend: Digitization through the electronic health record brings a patient’s entire clinical history to one place through notes, observations, lab results, imaging and other data. There is a risk that digitization creates only copies of the old analog data sets - electronic records without analytics and interoperability are just expensive paper. To truly see the advantage in digitization the data needs to be used for the purposes of better outcomes, population health or reducing waste and duplication. Digitization is also allowing for other more complex and relevant data from non-health sources to be compiled to create a holistic view of either a community or individual. Social determinants of health are being included more and more, not for healthcare but for health prevention too.
Prediction: We will see a move to increase algorithms that are classified and approved as software as a medical device. These are still very simple and don't offer a great deal of clinical insight but that is likely to change as the regulations around approval become clearer and self-certification is mapped out. We think of these processes like training a first-year medical student -- building the basics of medical knowledge and outcomes that will grow in complexity over time.
- We need to be realistic about AI and technology. Many administrative tasks are repetitive, redundant and do not require manual intervention, making them prime targets for AI applications. If AI advocates can move past the vision of the computer doing medicine and start with the building blocks of automating intelligence, there could be a giant leap forward in the space.
- Integration of non-healthcare data into analytic models will raise significant ethical issues and be a societal discussion around the values we have and how they translate to health. An example would be that if there is a care area that is not economically viable, that is way over budget and not producing the best outcomes do we stop funding it? Maybe the data says yes, but when that area is neonatal intensive care our ethics and values say no.
- Model bias and the ethics around prediction and automation of decisioning. That is why we often talk about augmented intelligence and supervised learning. We are looking to ensure that data and observational bias does not creep into the models.
About The Author
Greg Horne is the SAS Global Principal for Health and is based out of Toronto, Canada. Throughout his career, Greg has worked and become familiar with healthcare systems in the United Kingdom, Europe, North America and Australia. Greg began working as a Radiographer at the University College hospital in London, UK. He developed a passion for creating health systems built on quality and patient outcomes. He is considered a thought leader in the future of healthcare and the introduction of patient focused technology. As the Principal Consultant for Health Care, Greg has the opportunity to work with healthcare strategy in a way that focuses on outcomes as well as the cost, quality and other challenges that any modern health system faces. Greg graduated with a Bachelors in Radiography from the University of Southampton, UK. He is the recipient of a Canadian award for his work on using unstructured data to predict mental health issues with social media.