By Algorithms, machine learning, and artificial intelligence are slowly taking the world by storm, upending traditional processes across virtually every industry, and healthcare is no different. When you pull up your Netflix account, the website will recommend certain titles based on your previous viewing habits. Meteorologists use previous weather patterns to predict future outcomes. And now doctors can look to millions of electronic health records (EHR) when diagnosing and treating their patients.
Major tech companies like Google are trying to collect as much health data as they can to create a sort of online search tool for doctors and physicians. When a doctor evaluates a patient, previous clinical outcomes and patient data will tell them what to do next, so they can move on to the next patient as quickly as possible without second-guessing themselves or compromising patient care. If you’re curious about the future of healthcare, dive into all things AI and what we mean by “predictive care”.
The Poverty of Attention
Doctors have access to large quantities of data when evaluating their patients, including the patient’s health history, similar clinical outcomes, and information regarding the latest treatment methods. However, when we’re exposed to too much information all at once, it creates what’s known as “poverty of attention”. Our minds simply can’t process and absorb this information in a meaningful way. Part of Google’s plan is to condense healthcare information as much as possible, so doctors and physicians don’t have to read through pages of data when treating their patients.
Making Decisions on the Spot
Every second counts when a patient’s life is on the line, so doctors need to work fast. Doctors are also working harder than ever before, and they need to use their time wisely if they’re going to treat all their patients effectively.
According to a 2018 survey by the Physicians Foundation, doctors on average work 51 hours a week and see 20 patients a day. Considering their workload, doctors need to be able to access meaningful insights in a matter of seconds instead of combing through a patient’s entire medical history.
At any given moment, doctors need to decide whom they’re going to treat first and how based on data from their chart. Deciding which data to focus on and how to best treat the patient isn’t always easy. With AI-assisted technology, the algorithm can start making some of these decisions on the doctor’s behalf, so they can spend more time doing the work and less time figuring out what to do next.
Aggregate Electronic Health Records
To help doctors make complex healthcare decisions on the spot, Google is in the process of collecting a wealth of healthcare information from a diverse array of sources. To comply with patient privacy laws, such as those outlined in HIPAA, hospitals have agreed to de-identify patients before releasing these records. The EHRs contain a range of important information, including medications, laboratory values, diagnoses, vital signs, and medical notes.
All these EHRs will then be converted into a single standardized data structure format, with relevant information ordered and arranged per individual patient. Google’s new invention will then use three deep learning methods to predict future clinical outcomes. The technology will also be able to quickly summarize past medical events and incidents with patients with similar conditions as they relate to the situation at hand. Using this model, doctors could see the final outcome before ever treating the patient.
The tool will be able to predict a range of important clinical information, including:
- Unplanned transfers to the intensive care unit
- Unplanned hospitalizations
- ER visits or readmissions within 30 days after discharging the patient
- Patient length of stay in the hospital
- Inpatient mortality
- Primary diagnosis
- Atypical laboratory results that could signal a new chronic condition or health issue
- Primary and secondary billing diagnoses at patient discharge
Doctors will receive an alert on their tablet or smartphone when the machine predicts future clinical events. The alert will also contain “attention mechanisms” that track how much information the machine accessed when predicting the outcome, such as the number of individual words in a note, lab measurements, medications, etc. This helps the doctor decide how much confidence they should place in the predicted outcome.
It’s important to remember that this new machine is meant to be used as a tool. Doctors will be free to use or refuse information regarding predicted outcomes as they see fit. AI algorithms will only supplement their medical knowledge and expertise, not replace it entirely. While this project is still in the early phases, Google has focused its efforts on ophthalmology and digital pathology. The more information Google collects over the years, the more comprehensive the algorithm will be.
Google isn’t the only one experimenting with predictive healthcare. Amazon is looking upload EHRs to the cloud, while researchers at the University of Nottingham recently created an AI algorithm that can predict death. Depending on the results of this initial experiment, AI and healthcare may prove to be a match made in heaven.
